DEMOThis report is public and visible to anyone with the link for 30 days, then deleted. Save privately — create free account
gemma-4-E4B-it-SDFT_Heretic_RP.i1-IQ3_M.gguf
Format: GGUF  ·  Size: 4496.3 MB  ·  Scanned: 2026-09-15  ·  82480 ms
SHA256: 434ed4a3938649780c23486849e2d62138d681e3…

Security grade

B
Score: 81/100
2 MEDIUM 1 LOW 5 INFO
Artifact Security
B / 81
Repository Trust
50 / 100
Deployment Confidence
Medium
Artifact Security grades the file. Repository Trust reflects HuggingFace repo metadata signals (license, README, popularity). Deployment Confidence combines both.
▶ Score breakdown
Base score 100
MEDIUM Complex agentic chat template — context-dependent risk -8
MEDIUM External URLs embedded in 3 KV metadata field(s) -8
LOW 134 weight tensor(s) are >20x the median size: per_layer_model_proj.weight, blk.0.ffn_down.weight, blk.0.ffn_gate.weight -3
Penalty subtotal 81/100
Final B / 81
GGUF v3 model | architecture: gemma4 | size: 7.5B | quantization: Q6_K | 720 tensors | 3 warnings / 5 informational

Runtime compatibility

Runtime Status Quant Architecture Context
llama.cpp
Verify arch support
 ✓ ✓ ?
verify
 128k — may require flash attention (--flash-attn) and large VRAM
Ollama
Verify arch support
 ✓ ✓ ?
verify
 128k — may require flash attention (--flash-attn) and large VRAM
LM Studio
Verify arch support
 ✓ ✓ ?
verify
 128k — may require flash attention (--flash-attn) and large VRAM
vLLM ✗ N/A GGUF not supported — requires safetensors format
HuggingFace Transformers ✗ N/A GGUF not directly supported — use safetensors or PyTorch checkpoint

Model info

Architecture
Neural network family — determines which runtime can load this model
gemma4
NameGemma 4 E4B It SDFT_Heretic_RP
Size label
Human-readable parameter count from model metadata
7.5B
Quantization
Weight storage format derived from actual tensor dtypes; lower bits = smaller file and faster inference at the cost of accuracy
 Q6_K
ⓘ Declared general.file_type: IQ3_S — does not match actual tensors (companion/projector file likely inheriting main model metadata)
Parameters ~
Estimated total number of weight values; determines VRAM needed at inference
7,518,069,290
Context length
Maximum tokens the model can process in a single prompt+response; affects KV-cache memory
131,072 tokens
Embedding dim
Hidden state dimension (d_model); larger = more expressive but more compute per token
2,560
FFN dim
Feed-forward network intermediate size; typically 2.7–4× embedding_length
10,240
Layers
Number of transformer blocks (depth); more layers = more reasoning capacity
42
Attention heads (Q)
Number of query heads; GQA/MQA models use fewer KV heads than Q heads
8
KV heads
Key/Value heads per layer; fewer than Q heads = GQA; list = per-layer (hybrid architecture)
 2 (GQA: 8 Q → 2 KV)
Vocab size
Number of unique tokens the tokenizer knows; affects embedding table size
262,144
BOS / EOS tokens
Beginning-of-sequence and end-of-sequence token IDs used by the tokenizer
2 / 1
Tokenizer
Tokenizer algorithm family (e.g. gpt2 = BPE, llama = SentencePiece)
gemma4
Chat template
Jinja2 prompt template embedded in model; controls how messages are formatted for inference
✓ Yes — see section below
Tensors
Total number of weight tensors stored in the file
720
File size
Size on disk in megabytes
4496.28 MB
License
Usage license declared in model metadata; check terms before commercial deployment
apache-2.0

Security findings (8)

MEDIUM Metadata Security
Complex agentic chat template — context-dependent risk
Template uses tool_calls, tool responses, namespace() — the full pattern for agentic pipelines. These are legitimate HuggingFace constructs but represent the highest-complexity template class. Risk is context-dependent: negligible for static inference, meaningful in autonomous-agent deployments where user-controlled content can reach the template renderer.
→ Review template before use in agentic pipelines. Use sandboxed Jinja2 rendering and compare against the upstream author's reference template.
MEDIUM Metadata Security
External URLs embedded in 3 KV metadata field(s)
Fields: ['general.base_model.0.repo_url', 'general.url', 'general.source.url']
→ URLs in model metadata may be used for tracking or exfiltration when metadata is rendered; do not auto-fetch
LOW Statistical Anomaly
134 weight tensor(s) are >20x the median size: per_layer_model_proj.weight, blk.0.ffn_down.weight, blk.0.ffn_gate.weight
Extreme outlier tensors may contain embedded payloads or crafted weight matrices
→ Inspect outlier tensor names and compare against reference architecture
INFO Architecture
Grouped Query Attention (GQA) detected
head_count=8, head_count_kv=2 — requires GQA-aware runtime
→ Ensure your runtime (llama.cpp >= b1500, Ollama >= 0.1.30) supports GQA
INFO Embedding Size
Large embedding tensor 'per_layer_token_embd.weight': 2,312,110,080 bytes
vocab=262,144 x emb_dim=10,752 x 0.8203125 B/elem -> expected 2,312,110,080 B — size is consistent with architecture
→ No action needed; large embeddings are expected for large-vocabulary models
INFO Embedding Size
Large embedding tensor 'token_embd.weight': 550,502,400 bytes
vocab=262,144 x emb_dim=2,560 x 0.8203125 B/elem -> expected 550,502,400 B — size is consistent with architecture
→ No action needed; large embeddings are expected for large-vocabulary models
INFO Quantization
Model is quantized: IQ3_S, Q4_K, Q6_K
Quantization reduces size and inference cost at the expense of some accuracy
→ Verify the quantization level is appropriate for your use case (Q4_K_M is a common balanced choice)
INFO Quantization
Declared file_type (IQ3_S) does not match effective tensor dtypes (Q6_K)
The GGUF general.file_type field declares 'IQ3_S' but the actual tensors in this file are 'Q6_K'. This is common for multi-file model packages where the projector/adapter GGUF inherits the main model's file_type metadata. The effective dtype shown in the report reflects actual tensor storage.
→ No action required if this is a companion file (mmproj, adapter). For standalone models, this may indicate incorrect GGUF export.

Integrity & Structure

SHA-256434ed4a3938649780c23486849e2d62138d681e3628aa9396a1c83b0d5821f49
SHA-512094b9356abfe1d657d62f3df1bfc2ce340b6534e5b7739bf4fc5e81004081ede25c1ab15fd173f0a…
File size (bytes)4,714,691,264
Header + KV + tensor directory15,820,160 bytes (15.09 MB)
Tensor data region4,698,871,104 bytes (4481.19 MB)
Trailing bytes0
Unaccounted bytes0 ✓
KV metadata keys59
Tensors parsed720
Offsets monotone✓ OK
Max tensor offset4,698,589,504

DType Mix

DTypeCountTotal bytesTotal MBAvg bits/elem
Q6_K 2 2,862,612,480 2730.0 6.5625
IQ3_S 289 1,524,582,400 1453.96 3.4375
Q4_K 89 254,361,600 242.58 4.5
F16 1 55,050,240 52.5 16.0
F32 339 2,264,384 2.16 32.016628

Chat Template

tokenizer.chat_template (11926 chars) — click to collapse 
{%- macro format_parameters(properties, required) -%}
    {%- set standard_keys = ['description', 'type', 'properties', 'required', 'nullable'] -%}
    {%- set ns = namespace(found_first=false) -%}
    {%- for key, value in properties | dictsort -%}
        {%- set add_comma = false -%}
        {%- if key not in standard_keys -%}
            {%- if ns.found_first %},{% endif -%}
            {%- set ns.found_first = true -%}
            {{ key }}:{
            {%- if value['description'] -%}
                description:<|"|>{{ value['description'] }}<|"|>
                {%- set add_comma = true -%}
            {%- endif -%}
            {%- if value['nullable'] %}
                {%- if add_comma %},{%- else -%} {%- set add_comma = true -%} {% endif -%}
                nullable:true
            {%- endif -%}
            {%- if value['type'] | upper == 'STRING' -%}
                {%- if value['enum'] -%}
                    {%- if add_comma %},{%- else -%} {%- set add_comma = true -%} {% endif -%}
                    enum:{{ format_argument(value['enum']) }}
                {%- endif -%}
            {%- elif value['type'] | upper == 'OBJECT' -%}
                ,properties:{
                {%- if value['properties'] is defined and value['properties'] is mapping -%}
                    {{- format_parameters(value['properties'], value['required'] | default([])) -}}
                {%- elif value is mapping -%}
                    {{- format_parameters(value, value['required'] | default([])) -}}
                {%- endif -%}
                }
                {%- if value['required'] -%}
                    ,required:[
                    {%- for item in value['required'] | default([]) -%}
                        <|"|>{{- item -}}<|"|>
                        {%- if not loop.last %},{% endif -%}
                    {%- endfor -%}
                    ]
                {%- endif -%}
            {%- elif value['type'] | upper == 'ARRAY' -%}
                {%- if value['items'] is mapping and value['items'] -%}
                    ,items:{
                    {%- set ns_items = namespace(found_first=false) -%}
                    {%- for item_key, item_value in value['items'] | dictsort -%}
                        {%- if item_value is not none -%}
                            {%- if ns_items.found_first %},{% endif -%}
                            {%- set ns_items.found_first = true -%}
                            {%- if item_key == 'properties' -%}
                                properties:{
                                {%- if item_value is mapping -%}
                                    {{- format_parameters(item_value, value['items']['required'] | default([])) -}}
                                {%- endif -%}
                                }
                            {%- elif item_key == 'required' -%}
                                required:[
                                {%- for req_item in item_value -%}
                                    <|"|>{{- req_item -}}<|"|>
                                    {%- if not loop.last %},{% endif -%}
                                {%- endfor -%}
                                ]
                            {%- elif item_key == 'type' -%}
                                {%- if item_value is string -%}
                                    type:{{ format_argument(item_value | upper) }}
                                {%- else -%}
                                    type:{{ format_argument(item_value | map('upper') | list) }}
                                {%- endif -%}
                            {%- else -%}
                                {{ item_key }}:{{ format_argument(item_value) }}
                            {%- endif -%}
                        {%- endif -%}
                    {%- endfor -%}
                    }
                {%- endif -%}
            {%- endif -%}
            {%- if add_comma %},{%- else -%} {%- set add_comma = true -%} {% endif -%}
            type:<|"|>{{ value['type'] | upper }}<|"|>}
        {%- endif -%}
    {%- endfor -%}
{%- endmacro -%}
{%- macro format_function_declaration(tool_data) -%}
    declaration:{{- tool_data['function']['name'] -}}{description:<|"|>{{- tool_data['function']['description'] -}}<|"|>
    {%- set params = tool_data['function']['parameters'] -%}
    {%- if params -%}
        ,parameters:{
        {%- if params['properties'] -%}
            properties:{ {{- format_parameters(params['properties'], params['required']) -}} },
        {%- endif -%}
        {%- if params['required'] -%}
            required:[
            {%- for item in params['required'] -%}
                <|"|>{{- item -}}<|"|>
                {{- ',' if not loop.last -}}
            {%- endfor -%}
            ],
        {%- endif -%}
        {%- if params['type'] -%}
            type:<|"|>{{- params['type'] | upper -}}<|"|>}
        {%- endif -%}
    {%- endif -%}
    {%- if 'response' in tool_data['function'] -%}
        {%- set response_declaration = tool_data['function']['response'] -%}
        ,response:{
        {%- if response_declaration['description'] -%}
            description:<|"|>{{- response_declaration['description'] -}}<|"|>,
        {%- endif -%}
        {%- if response_declaration['type'] | upper == 'OBJECT' -%}
            type:<|"|>{{- response_declaration['type'] | upper -}}<|"|>}
        {%- endif -%}
    {%- endif -%}
    }
{%- endmacro -%}
{%- macro format_argument(argument, escape_keys=True) -%}
    {%- if argument is string -%}
        {{- '<|"|>' + argument + '<|"|>' -}}
    {%- elif argument is boolean -%}
        {{- 'true' if argument else 'false' -}}
    {%- elif argument is mapping -%}
        {{- '{' -}}
        {%- set ns = namespace(found_first=false) -%}
        {%- for key, value in argument | dictsort -%}
            {%- if ns.found_first %},{% endif -%}
            {%- set ns.found_first = true -%}
            {%- if escape_keys -%}
                {{- '<|"|>' + key + '<|"|>' -}}
            {%- else -%}
                {{- key -}}
            {%- endif -%}
            :{{- format_argument(value, escape_keys=escape_keys) -}}
        {%- endfor -%}
        {{- '}' -}}
    {%- elif argument is sequence -%}
        {{- '[' -}}
        {%- for item in argument -%}
            {{- format_argument(item, escape_keys=escape_keys) -}}
            {%- if not loop.last %},{% endif -%}
        {%- endfor -%}
        {{- ']' -}}
    {%- else -%}
        {{- argument -}}
    {%- endif -%}
{%- endmacro -%}
{%- macro strip_thinking(text) -%}
    {%- set ns = namespace(result='') -%}
    {%- for part in text.split('<channel|>') -%}
        {%- if '<|channel>' in part -%}
            {%- set ns.result = ns.result + part.split('<|channel>')[0] -%}
        {%- else -%}
            {%- set ns.result = ns.result + part -%}
        {%- endif -%}
    {%- endfor -%}
    {{- ns.result | trim -}}
{%- endmacro -%}

{%- set ns = namespace(prev_message_type=None) -%}
{%- set loop_messages = messages -%}
{{ bos_token }}
{#- Handle System/Tool Definitions Block -#}
{%- if (enable_thinking is defined and enable_thinking) or tools or messages[0]['role'] in ['system', 'developer'] -%}
    {{- '<|turn>system\n' -}}

    {#- Inject Thinking token at the very top of the FIRST system turn -#}
    {%- if enable_thinking is defined and enable_thinking -%}
        {{- '<|think|>' -}}
        {%- set ns.prev_message_type = 'think' -%}
    {%- endif -%}

    {%- if messages[0]['role'] in ['system', 'developer'] -%}
        {{- messages[0]['content'] | trim -}}
        {%- set loop_messages = messages[1:] -%}
    {%- endif -%}

    {%- if tools -%}
        {%- for tool in tools %}
            {{- '<|tool>' -}}
            {{- format_function_declaration(tool) | trim -}}
            {{- '<tool|>' -}}
        {%- endfor %}
        {%- set ns.prev_message_type = 'tool' -%}
    {%- endif -%}

    {{- '<turn|>\n' -}}
{%- endif %}

{#- Loop through messages -#}
{%- for message in loop_messages -%}
    {%- set ns.prev_message_type = None -%}
    {%- set role = 'model' if message['role'] == 'assistant' else message['role'] -%}
        {{- '<|turn>' + role + '\n' }}

            {%- if message['tool_calls'] -%}
                {%- for tool_call in message['tool_calls'] -%}
                    {%- set function = tool_call['function'] -%}
                    {{- '<|tool_call>call:' + function['name'] + '{' -}}
                    {%- if function['arguments'] is mapping -%}
                        {%- set ns_args = namespace(found_first=false) -%}
                        {%- for key, value in function['arguments'] | dictsort -%}
                            {%- if ns_args.found_first %},{% endif -%}
                            {%- set ns_args.found_first = true -%}
                            {{- key -}}:{{- format_argument(value, escape_keys=False) -}}
                        {%- endfor -%}
                    {%- elif function['arguments'] is string -%}
                        {{- function['arguments'] -}}
                    {%- endif -%}
                    {{- '}<tool_call|>' -}}
                {%- endfor -%}
                {%- set ns.prev_message_type = 'tool_call' -%}
            {%- endif -%}

            {%- if message['tool_responses'] -%}
                {#- Tool Response handling -#}
                {%- for tool_response in message['tool_responses'] -%}
                    {{- '<|tool_response>' -}}
                    {%- if tool_response['response'] is mapping -%}
                        {{- 'response:' + tool_response['name'] | default('unknown') + '{' -}}
                        {%- for key, value in tool_response['response'] | dictsort -%}
                            {{- key -}}:{{- format_argument(value, escape_keys=False) -}}
                            {%- if not loop.last %},{% endif -%}
                        {%- endfor -%}
                        {{- '}' -}}
                    {%- else -%}
                        {{- 'response:' + tool_response['name'] | default('unknown') + '{value:' + format_argument(tool_response['response'], escape_keys=False) + '}' -}}
                    {%- endif -%}
                    {{- '<tool_response|>' -}}
                {%- endfor -%}
                {%- set ns.prev_message_type = 'tool_response' -%}
            {%- endif -%}

            {%- if message['content'] is string -%}
                {%- if role == 'model' -%}
                    {{- strip_thinking(message['content']) -}}
                {%- else -%}
                    {{- message['content'] | trim -}}
                {%- endif -%}
            {%- elif message['content'] is sequence -%}
                {%- for item in message['content'] -%}
                    {%- if item['type'] == 'text' -%}
                        {%- if role == 'model' -%}
                            {{- strip_thinking(item['text']) -}}
                        {%- else -%}
                            {{- item['text'] | trim -}}
                        {%- endif -%}
                    {%- elif item['type'] == 'image' -%}
                        {{- '\n\n<|image|>\n\n' -}}
                        {%- set ns.prev_message_type = 'image' -%}
                    {%- elif item['type'] == 'audio' -%}
                        {{- '<|audio|>' -}}
                        {%- set ns.prev_message_type = 'audio' -%}
                    {%- elif item['type'] == 'video' -%}
                        {{- '\n\n<|video|>\n\n' -}}
                        {%- set ns.prev_message_type = 'video' -%}
                    {%- endif -%}
                {%- endfor -%}
            {%- endif -%}

        {%- if not (message['tool_responses'] and not message['content']) -%}
            {{- '<turn|>\n' -}}
        {%- endif -%}
{%- endfor -%}

{%- if add_generation_prompt -%}
    {%- if ns.prev_message_type != 'tool_response' -%}
        {{- '<|turn>model\n' -}}
    {%- endif -%}
{%- endif -%}

This is a prompt template (non-executable). If your application implements tool-calling agents, enforce a whitelist + schema validation for any <tool_call> output.

Tensor Preview (first 50 of 720)

#NameDimsDTypeOffsetBytesbits/elem
0 output_norm.weight [2560] F32 0 10,240 32.0
1 per_layer_model_proj.weight [2560, 10752] F16 10,240 55,050,240 16.0
2 per_layer_proj_norm.weight [256] F32 55,060,480 1,024 32.0
3 per_layer_token_embd.weight [10752, 262144] Q6_K 55,061,504 2,312,110,080 6.5625
4 rope_freqs.weight [256] F32 2,367,171,584 1,024 32.0
5 token_embd.weight [2560, 262144] Q6_K 2,367,172,608 550,502,400 6.5625
6 blk.0.attn_k.weight [2560, 512] IQ3_S 2,917,675,008 563,200 3.4375
7 blk.0.attn_k_norm.weight [256] F32 2,918,238,208 1,024 32.0
8 blk.0.attn_norm.weight [2560] F32 2,918,239,232 10,240 32.0
9 blk.0.attn_output.weight [2048, 2560] Q4_K 2,918,249,472 2,949,120 4.5
10 blk.0.attn_q.weight [2560, 2048] IQ3_S 2,921,198,592 2,252,800 3.4375
11 blk.0.attn_q_norm.weight [256] F32 2,923,451,392 1,024 32.0
12 blk.0.attn_v.weight [2560, 512] Q4_K 2,923,452,416 737,280 4.5
13 blk.0.ffn_down.weight [10240, 2560] Q4_K 2,924,189,696 14,745,600 4.5
14 blk.0.ffn_gate.weight [2560, 10240] IQ3_S 2,938,935,296 11,264,000 3.4375
15 blk.0.ffn_norm.weight [2560] F32 2,950,199,296 10,240 32.0
16 blk.0.ffn_up.weight [2560, 10240] IQ3_S 2,950,209,536 11,264,000 3.4375
17 blk.0.inp_gate.weight [2560, 256] IQ3_S 2,961,473,536 281,600 3.4375
18 blk.0.layer_output_scale.weight [1] F32 2,961,755,136 32 256.0
19 blk.0.post_attention_norm.weight [2560] F32 2,961,755,168 10,240 32.0
20 blk.0.post_ffw_norm.weight [2560] F32 2,961,765,408 10,240 32.0
21 blk.0.post_norm.weight [2560] F32 2,961,775,648 10,240 32.0
22 blk.0.proj.weight [256, 2560] IQ3_S 2,961,785,888 281,600 3.4375
23 blk.1.attn_k.weight [2560, 512] IQ3_S 2,962,067,488 563,200 3.4375
24 blk.1.attn_k_norm.weight [256] F32 2,962,630,688 1,024 32.0
25 blk.1.attn_norm.weight [2560] F32 2,962,631,712 10,240 32.0
26 blk.1.attn_output.weight [2048, 2560] Q4_K 2,962,641,952 2,949,120 4.5
27 blk.1.attn_q.weight [2560, 2048] IQ3_S 2,965,591,072 2,252,800 3.4375
28 blk.1.attn_q_norm.weight [256] F32 2,967,843,872 1,024 32.0
29 blk.1.attn_v.weight [2560, 512] Q4_K 2,967,844,896 737,280 4.5
30 blk.1.ffn_down.weight [10240, 2560] Q4_K 2,968,582,176 14,745,600 4.5
31 blk.1.ffn_gate.weight [2560, 10240] IQ3_S 2,983,327,776 11,264,000 3.4375
32 blk.1.ffn_norm.weight [2560] F32 2,994,591,776 10,240 32.0
33 blk.1.ffn_up.weight [2560, 10240] IQ3_S 2,994,602,016 11,264,000 3.4375
34 blk.1.inp_gate.weight [2560, 256] IQ3_S 3,005,866,016 281,600 3.4375
35 blk.1.layer_output_scale.weight [1] F32 3,006,147,616 32 256.0
36 blk.1.post_attention_norm.weight [2560] F32 3,006,147,648 10,240 32.0
37 blk.1.post_ffw_norm.weight [2560] F32 3,006,157,888 10,240 32.0
38 blk.1.post_norm.weight [2560] F32 3,006,168,128 10,240 32.0
39 blk.1.proj.weight [256, 2560] IQ3_S 3,006,178,368 281,600 3.4375
40 blk.2.attn_k.weight [2560, 512] IQ3_S 3,006,459,968 563,200 3.4375
41 blk.2.attn_k_norm.weight [256] F32 3,007,023,168 1,024 32.0
42 blk.2.attn_norm.weight [2560] F32 3,007,024,192 10,240 32.0
43 blk.2.attn_output.weight [2048, 2560] Q4_K 3,007,034,432 2,949,120 4.5
44 blk.2.attn_q.weight [2560, 2048] IQ3_S 3,009,983,552 2,252,800 3.4375
45 blk.2.attn_q_norm.weight [256] F32 3,012,236,352 1,024 32.0
46 blk.2.attn_v.weight [2560, 512] Q4_K 3,012,237,376 737,280 4.5
47 blk.2.ffn_down.weight [10240, 2560] Q4_K 3,012,974,656 14,745,600 4.5
48 blk.2.ffn_gate.weight [2560, 10240] IQ3_S 3,027,720,256 11,264,000 3.4375
49 blk.2.ffn_norm.weight [2560] F32 3,038,984,256 10,240 32.0
… 670 more tensors omitted from preview

HuggingFace Provenance

Repositorymradermacher/gemma-4-E4B-it-SDFT_Heretic_RP-i1-GGUF
File pathgemma-4-E4B-it-SDFT_Heretic_RP.i1-IQ3_M.gguf
Repo commita6356be002df20b7c3abaa96b90b0ff328e2b0cf
License (model card)apache-2.0
README.md✓ present
config.json— not found
tokenizer_config.json— not found
Siblings in repo18 file(s)
Downloads1,156
Likes4
Repo trust score
Heuristic 0–100: license + README + config + download/like signals
 50/100
Tagstransformers, gguf, gemma4, roleplay, rp, merged, sdft, gguf-ready, en, base_model:Ilya626/gemma-4-E4B-it-SDFT_Heretic_RP…
All files in repository (18)
FileSize
.gitattributes 0.0 MB
README.md 0.0 MB
gemma-4-E4B-it-SDFT_Heretic_RP.i1-IQ3_M.gguf 4496.3 MB
gemma-4-E4B-it-SDFT_Heretic_RP.i1-IQ3_S.gguf 4447.1 MB
gemma-4-E4B-it-SDFT_Heretic_RP.i1-IQ4_NL.gguf 4953.3 MB
gemma-4-E4B-it-SDFT_Heretic_RP.i1-IQ4_XS.gguf 4836.0 MB
gemma-4-E4B-it-SDFT_Heretic_RP.i1-Q2_K.gguf 4197.4 MB
gemma-4-E4B-it-SDFT_Heretic_RP.i1-Q3_K_L.gguf 4788.7 MB
gemma-4-E4B-it-SDFT_Heretic_RP.i1-Q3_K_M.gguf 4625.7 MB
gemma-4-E4B-it-SDFT_Heretic_RP.i1-Q3_K_S.gguf 4439.0 MB
gemma-4-E4B-it-SDFT_Heretic_RP.i1-Q4_0.gguf 4953.5 MB
gemma-4-E4B-it-SDFT_Heretic_RP.i1-Q4_1.gguf 5184.1 MB
gemma-4-E4B-it-SDFT_Heretic_RP.i1-Q4_K_M.gguf 5088.1 MB
gemma-4-E4B-it-SDFT_Heretic_RP.i1-Q4_K_S.gguf 4961.9 MB
gemma-4-E4B-it-SDFT_Heretic_RP.i1-Q5_K_M.gguf 5495.9 MB
gemma-4-E4B-it-SDFT_Heretic_RP.i1-Q5_K_S.gguf 5422.6 MB
gemma-4-E4B-it-SDFT_Heretic_RP.i1-Q6_K.gguf 5929.2 MB
gemma-4-E4B-it-SDFT_Heretic_RP.imatrix.gguf 4.2 MB

Scan coverage

Checks performed
  • ✓Magic bytes &amp; GGUF version validation
  • ✓Header parse (tensor count, KV count)
  • ✓KV metadata parse — all key/value pairs
  • ✓Tensor directory parse (names, dims, dtype, offsets)
  • ✓Tensor offset monotonicity
  • ✓Tensor overlap detection
  • ✓Tensor out-of-bounds detection
  • ✓Data section alignment verification
  • ✓Trailing bytes after last tensor
  • ✓Unaccounted bytes (header vs file size)
  • ✓Dtype histogram &amp; quantization mix analysis
  • ✓Embedding tensor size cross-validation
  • ✓Chat template static scan (Jinja2/SSTI patterns)
  • ✓Runtime compatibility heuristics
  • ✓HuggingFace model card & repository context checked
  • ✓Upstream HuggingFace repository metadata checked
  • ✓Repo identity & commit SHA captured (a6356be0…)
  • ✓README.md presence checked (✓ present)
  • ✓config.json / tokenizer_config.json presence checked
  • ✓Sibling files enumerated (18 file(s))
  • ✓Downloads / likes / tags captured
  • —Full repository file content audit: not performed
  • —License legal text validation: not performed
Not covered
  • ✗Runtime execution or dynamic analysis
  • ✗Behavioral backdoor detection
  • ✗Weight-level semantic backdoor detection
  • ✗Full upstream repository verification
  • ✗License validation beyond embedded metadata
  • ✗Malware scanning of surrounding repo files
  • ✗Adversarial robustness or alignment audit
  • ✗Training data provenance
ⓘ Static scanning can detect structural anomalies and known malicious patterns. It cannot guarantee absence of risk. Treat results as a security signal, not a formal audit.
Raw scan data (.llmscan JSON) 
{
  "compatibility": [
    {
      "arch": null,
      "container": true,
      "context": "128k \u2014 may require flash attention (--flash-attn) and large VRAM",
      "notes": "GGUF v3 container: supported | quant Q6_K: supported | 128k context: requires sufficient VRAM/RAM and --ctx-size flag \u2014 architecture \u0027gemma4\u0027 is uncommon; check runtime release notes for explicit support",
      "quant": true,
      "runtime": "llama.cpp",
      "supported": null
    },
    {
      "arch": null,
      "container": true,
      "context": "128k \u2014 may require flash attention (--flash-attn) and large VRAM",
      "notes": "GGUF via llama.cpp backend: container and quantization supported \u2014 architecture \u0027gemma4\u0027 is uncommon; check runtime release notes for explicit support",
      "quant": true,
      "runtime": "Ollama",
      "supported": null
    },
    {
      "arch": null,
      "container": true,
      "context": "128k \u2014 may require flash attention (--flash-attn) and large VRAM",
      "notes": "GGUF container and quantization supported; GPU/CPU inference \u2014 architecture \u0027gemma4\u0027 is uncommon; check runtime release notes for explicit support",
      "quant": true,
      "runtime": "LM Studio",
      "supported": null
    },
    {
      "arch": null,
      "container": false,
      "context": null,
      "not_supported_msg": "GGUF not supported \u2014 requires safetensors format",
      "notes": "GGUF not supported; requires safetensors format",
      "quant": false,
      "runtime": "vLLM",
      "supported": false
    },
    {
      "arch": null,
      "container": false,
      "context": null,
      "not_supported_msg": "GGUF not directly supported \u2014 use safetensors or PyTorch checkpoint",
      "notes": "GGUF not supported; requires safetensors or PyTorch checkpoint",
      "quant": false,
      "runtime": "HuggingFace Transformers",
      "supported": false
    }
  ],
  "format": "LLM-SCAN",
  "format_version": "1.0",
  "generator": {
    "name": "llmscan-engine",
    "version": "2.0.0"
  },
  "model": {
    "alignment": 32,
    "arch": "gemma4",
    "architecture": "gemma4",
    "basename": null,
    "block_count": 42,
    "bos_token_id": 2,
    "chat_template": "{%- macro format_parameters(properties, required) -%}\n    {%- set standard_keys = [\u0027description\u0027, \u0027type\u0027, \u0027properties\u0027, \u0027required\u0027, \u0027nullable\u0027] -%}\n    {%- set ns = namespace(found_first=false) -%}\n    {%- for key, value in properties | dictsort -%}\n        {%- set add_comma = false -%}\n        {%- if key not in standard_keys -%}\n            {%- if ns.found_first %},{% endif -%}\n            {%- set ns.found_first = true -%}\n            {{ key }}:{\n            {%- if value[\u0027description\u0027] -%}\n                description:\u003c|\"|\u003e{{ value[\u0027description\u0027] }}\u003c|\"|\u003e\n                {%- set add_comma = true -%}\n            {%- endif -%}\n            {%- if value[\u0027nullable\u0027] %}\n                {%- if add_comma %},{%- else -%} {%- set add_comma = true -%} {% endif -%}\n                nullable:true\n            {%- endif -%}\n            {%- if value[\u0027type\u0027] | upper == \u0027STRING\u0027 -%}\n                {%- if value[\u0027enum\u0027] -%}\n                    {%- if add_comma %},{%- else -%} {%- set add_comma = true -%} {% endif -%}\n                    enum:{{ format_argument(value[\u0027enum\u0027]) }}\n                {%- endif -%}\n            {%- elif value[\u0027type\u0027] | upper == \u0027OBJECT\u0027 -%}\n                ,properties:{\n                {%- if value[\u0027properties\u0027] is defined and value[\u0027properties\u0027] is mapping -%}\n                    {{- format_parameters(value[\u0027properties\u0027], value[\u0027required\u0027] | default([])) -}}\n                {%- elif value is mapping -%}\n                    {{- format_parameters(value, value[\u0027required\u0027] | default([])) -}}\n                {%- endif -%}\n                }\n                {%- if value[\u0027required\u0027] -%}\n                    ,required:[\n                    {%- for item in value[\u0027required\u0027] | default([]) -%}\n                        \u003c|\"|\u003e{{- item -}}\u003c|\"|\u003e\n                        {%- if not loop.last %},{% endif -%}\n                    {%- endfor -%}\n                    ]\n                {%- endif -%}\n            {%- elif value[\u0027type\u0027] | upper == \u0027ARRAY\u0027 -%}\n                {%- if value[\u0027items\u0027] is mapping and value[\u0027items\u0027] -%}\n                    ,items:{\n                    {%- set ns_items = namespace(found_first=false) -%}\n                    {%- for item_key, item_value in value[\u0027items\u0027] | dictsort -%}\n                        {%- if item_value is not none -%}\n                            {%- if ns_items.found_first %},{% endif -%}\n                            {%- set ns_items.found_first = true -%}\n                            {%- if item_key == \u0027properties\u0027 -%}\n                                properties:{\n                                {%- if item_value is mapping -%}\n                                    {{- format_parameters(item_value, value[\u0027items\u0027][\u0027required\u0027] | default([])) -}}\n                                {%- endif -%}\n                                }\n                            {%- elif item_key == \u0027required\u0027 -%}\n                                required:[\n                                {%- for req_item in item_value -%}\n                                    \u003c|\"|\u003e{{- req_item -}}\u003c|\"|\u003e\n                                    {%- if not loop.last %},{% endif -%}\n                                {%- endfor -%}\n                                ]\n                            {%- elif item_key == \u0027type\u0027 -%}\n                                {%- if item_value is string -%}\n                                    type:{{ format_argument(item_value | upper) }}\n                                {%- else -%}\n                                    type:{{ format_argument(item_value | map(\u0027upper\u0027) | list) }}\n                                {%- endif -%}\n                            {%- else -%}\n                                {{ item_key }}:{{ format_argument(item_value) }}\n                            {%- endif -%}\n                        {%- endif -%}\n                    {%- endfor -%}\n                    }\n                {%- endif -%}\n            {%- endif -%}\n            {%- if add_comma %},{%- else -%} {%- set add_comma = true -%} {% endif -%}\n            type:\u003c|\"|\u003e{{ value[\u0027type\u0027] | upper }}\u003c|\"|\u003e}\n        {%- endif -%}\n    {%- endfor -%}\n{%- endmacro -%}\n{%- macro format_function_declaration(tool_data) -%}\n    declaration:{{- tool_data[\u0027function\u0027][\u0027name\u0027] -}}{description:\u003c|\"|\u003e{{- tool_data[\u0027function\u0027][\u0027description\u0027] -}}\u003c|\"|\u003e\n    {%- set params = tool_data[\u0027function\u0027][\u0027parameters\u0027] -%}\n    {%- if params -%}\n        ,parameters:{\n        {%- if params[\u0027properties\u0027] -%}\n            properties:{ {{- format_parameters(params[\u0027properties\u0027], params[\u0027required\u0027]) -}} },\n        {%- endif -%}\n        {%- if params[\u0027required\u0027] -%}\n            required:[\n            {%- for item in params[\u0027required\u0027] -%}\n                \u003c|\"|\u003e{{- item -}}\u003c|\"|\u003e\n                {{- \u0027,\u0027 if not loop.last -}}\n            {%- endfor -%}\n            ],\n        {%- endif -%}\n        {%- if params[\u0027type\u0027] -%}\n            type:\u003c|\"|\u003e{{- params[\u0027type\u0027] | upper -}}\u003c|\"|\u003e}\n        {%- endif -%}\n    {%- endif -%}\n    {%- if \u0027response\u0027 in tool_data[\u0027function\u0027] -%}\n        {%- set response_declaration = tool_data[\u0027function\u0027][\u0027response\u0027] -%}\n        ,response:{\n        {%- if response_declaration[\u0027description\u0027] -%}\n            description:\u003c|\"|\u003e{{- response_declaration[\u0027description\u0027] -}}\u003c|\"|\u003e,\n        {%- endif -%}\n        {%- if response_declaration[\u0027type\u0027] | upper == \u0027OBJECT\u0027 -%}\n            type:\u003c|\"|\u003e{{- response_declaration[\u0027type\u0027] | upper -}}\u003c|\"|\u003e}\n        {%- endif -%}\n    {%- endif -%}\n    }\n{%- endmacro -%}\n{%- macro format_argument(argument, escape_keys=True) -%}\n    {%- if argument is string -%}\n        {{- \u0027\u003c|\"|\u003e\u0027 + argument + \u0027\u003c|\"|\u003e\u0027 -}}\n    {%- elif argument is boolean -%}\n        {{- \u0027true\u0027 if argument else \u0027false\u0027 -}}\n    {%- elif argument is mapping -%}\n        {{- \u0027{\u0027 -}}\n        {%- set ns = namespace(found_first=false) -%}\n        {%- for key, value in argument | dictsort -%}\n            {%- if ns.found_first %},{% endif -%}\n            {%- set ns.found_first = true -%}\n            {%- if escape_keys -%}\n                {{- \u0027\u003c|\"|\u003e\u0027 + key + \u0027\u003c|\"|\u003e\u0027 -}}\n            {%- else -%}\n                {{- key -}}\n            {%- endif -%}\n            :{{- format_argument(value, escape_keys=escape_keys) -}}\n        {%- endfor -%}\n        {{- \u0027}\u0027 -}}\n    {%- elif argument is sequence -%}\n        {{- \u0027[\u0027 -}}\n        {%- for item in argument -%}\n            {{- format_argument(item, escape_keys=escape_keys) -}}\n            {%- if not loop.last %},{% endif -%}\n        {%- endfor -%}\n        {{- \u0027]\u0027 -}}\n    {%- else -%}\n        {{- argument -}}\n    {%- endif -%}\n{%- endmacro -%}\n{%- macro strip_thinking(text) -%}\n    {%- set ns = namespace(result=\u0027\u0027) -%}\n    {%- for part in text.split(\u0027\u003cchannel|\u003e\u0027) -%}\n        {%- if \u0027\u003c|channel\u003e\u0027 in part -%}\n            {%- set ns.result = ns.result + part.split(\u0027\u003c|channel\u003e\u0027)[0] -%}\n        {%- else -%}\n            {%- set ns.result = ns.result + part -%}\n        {%- endif -%}\n    {%- endfor -%}\n    {{- ns.result | trim -}}\n{%- endmacro -%}\n\n{%- set ns = namespace(prev_message_type=None) -%}\n{%- set loop_messages = messages -%}\n{{ bos_token }}\n{#- Handle System/Tool Definitions Block -#}\n{%- if (enable_thinking is defined and enable_thinking) or tools or messages[0][\u0027role\u0027] in [\u0027system\u0027, \u0027developer\u0027] -%}\n    {{- \u0027\u003c|turn\u003esystem\\n\u0027 -}}\n\n    {#- Inject Thinking token at the very top of the FIRST system turn -#}\n    {%- if enable_thinking is defined and enable_thinking -%}\n        {{- \u0027\u003c|think|\u003e\u0027 -}}\n        {%- set ns.prev_message_type = \u0027think\u0027 -%}\n    {%- endif -%}\n\n    {%- if messages[0][\u0027role\u0027] in [\u0027system\u0027, \u0027developer\u0027] -%}\n        {{- messages[0][\u0027content\u0027] | trim -}}\n        {%- set loop_messages = messages[1:] -%}\n    {%- endif -%}\n\n    {%- if tools -%}\n        {%- for tool in tools %}\n            {{- \u0027\u003c|tool\u003e\u0027 -}}\n            {{- format_function_declaration(tool) | trim -}}\n            {{- \u0027\u003ctool|\u003e\u0027 -}}\n        {%- endfor %}\n        {%- set ns.prev_message_type = \u0027tool\u0027 -%}\n    {%- endif -%}\n\n    {{- \u0027\u003cturn|\u003e\\n\u0027 -}}\n{%- endif %}\n\n{#- Loop through messages -#}\n{%- for message in loop_messages -%}\n    {%- set ns.prev_message_type = None -%}\n    {%- set role = \u0027model\u0027 if message[\u0027role\u0027] == \u0027assistant\u0027 else message[\u0027role\u0027] -%}\n        {{- \u0027\u003c|turn\u003e\u0027 + role + \u0027\\n\u0027 }}\n\n            {%- if message[\u0027tool_calls\u0027] -%}\n                {%- for tool_call in message[\u0027tool_calls\u0027] -%}\n                    {%- set function = tool_call[\u0027function\u0027] -%}\n                    {{- \u0027\u003c|tool_call\u003ecall:\u0027 + function[\u0027name\u0027] + \u0027{\u0027 -}}\n                    {%- if function[\u0027arguments\u0027] is mapping -%}\n                        {%- set ns_args = namespace(found_first=false) -%}\n                        {%- for key, value in function[\u0027arguments\u0027] | dictsort -%}\n                            {%- if ns_args.found_first %},{% endif -%}\n                            {%- set ns_args.found_first = true -%}\n                            {{- key -}}:{{- format_argument(value, escape_keys=False) -}}\n                        {%- endfor -%}\n                    {%- elif function[\u0027arguments\u0027] is string -%}\n                        {{- function[\u0027arguments\u0027] -}}\n                    {%- endif -%}\n                    {{- \u0027}\u003ctool_call|\u003e\u0027 -}}\n                {%- endfor -%}\n                {%- set ns.prev_message_type = \u0027tool_call\u0027 -%}\n            {%- endif -%}\n\n            {%- if message[\u0027tool_responses\u0027] -%}\n                {#- Tool Response handling -#}\n                {%- for tool_response in message[\u0027tool_responses\u0027] -%}\n                    {{- \u0027\u003c|tool_response\u003e\u0027 -}}\n                    {%- if tool_response[\u0027response\u0027] is mapping -%}\n                        {{- \u0027response:\u0027 + tool_response[\u0027name\u0027] | default(\u0027unknown\u0027) + \u0027{\u0027 -}}\n                        {%- for key, value in tool_response[\u0027response\u0027] | dictsort -%}\n                            {{- key -}}:{{- format_argument(value, escape_keys=False) -}}\n                            {%- if not loop.last %},{% endif -%}\n                        {%- endfor -%}\n                        {{- \u0027}\u0027 -}}\n                    {%- else -%}\n                        {{- \u0027response:\u0027 + tool_response[\u0027name\u0027] | default(\u0027unknown\u0027) + \u0027{value:\u0027 + format_argument(tool_response[\u0027response\u0027], escape_keys=False) + \u0027}\u0027 -}}\n                    {%- endif -%}\n                    {{- \u0027\u003ctool_response|\u003e\u0027 -}}\n                {%- endfor -%}\n                {%- set ns.prev_message_type = \u0027tool_response\u0027 -%}\n            {%- endif -%}\n\n            {%- if message[\u0027content\u0027] is string -%}\n                {%- if role == \u0027model\u0027 -%}\n                    {{- strip_thinking(message[\u0027content\u0027]) -}}\n                {%- else -%}\n                    {{- message[\u0027content\u0027] | trim -}}\n                {%- endif -%}\n            {%- elif message[\u0027content\u0027] is sequence -%}\n                {%- for item in message[\u0027content\u0027] -%}\n                    {%- if item[\u0027type\u0027] == \u0027text\u0027 -%}\n                        {%- if role == \u0027model\u0027 -%}\n                            {{- strip_thinking(item[\u0027text\u0027]) -}}\n                        {%- else -%}\n                            {{- item[\u0027text\u0027] | trim -}}\n                        {%- endif -%}\n                    {%- elif item[\u0027type\u0027] == \u0027image\u0027 -%}\n                        {{- \u0027\\n\\n\u003c|image|\u003e\\n\\n\u0027 -}}\n                        {%- set ns.prev_message_type = \u0027image\u0027 -%}\n                    {%- elif item[\u0027type\u0027] == \u0027audio\u0027 -%}\n                        {{- \u0027\u003c|audio|\u003e\u0027 -}}\n                        {%- set ns.prev_message_type = \u0027audio\u0027 -%}\n                    {%- elif item[\u0027type\u0027] == \u0027video\u0027 -%}\n                        {{- \u0027\\n\\n\u003c|video|\u003e\\n\\n\u0027 -}}\n                        {%- set ns.prev_message_type = \u0027video\u0027 -%}\n                    {%- endif -%}\n                {%- endfor -%}\n            {%- endif -%}\n\n        {%- if not (message[\u0027tool_responses\u0027] and not message[\u0027content\u0027]) -%}\n            {{- \u0027\u003cturn|\u003e\\n\u0027 -}}\n        {%- endif -%}\n{%- endfor -%}\n\n{%- if add_generation_prompt -%}\n    {%- if ns.prev_message_type != \u0027tool_response\u0027 -%}\n        {{- \u0027\u003c|turn\u003emodel\\n\u0027 -}}\n    {%- endif -%}\n{%- endif -%}",
    "context_length": 131072,
    "dtype_histogram": {
      "F16": 1,
      "F32": 339,
      "IQ3_S": 289,
      "Q4_K": 89,
      "Q6_K": 2
    },
    "dtype_percentage": {
      "F16": 0.1,
      "F32": 47.1,
      "IQ3_S": 40.1,
      "Q4_K": 12.4,
      "Q6_K": 0.3
    },
    "dtype_stats": {
      "F16": {
        "avg_bits_per_elem": 16.0,
        "bytes": 55050240,
        "count": 1,
        "mb": 52.5
      },
      "F32": {
        "avg_bits_per_elem": 32.016628,
        "bytes": 2264384,
        "count": 339,
        "mb": 2.16
      },
      "IQ3_S": {
        "avg_bits_per_elem": 3.4375,
        "bytes": 1524582400,
        "count": 289,
        "mb": 1453.96
      },
      "Q4_K": {
        "avg_bits_per_elem": 4.5,
        "bytes": 254361600,
        "count": 89,
        "mb": 242.58
      },
      "Q6_K": {
        "avg_bits_per_elem": 6.5625,
        "bytes": 2862612480,
        "count": 2,
        "mb": 2730.0
      }
    },
    "embedding_length": 2560,
    "eos_token_id": 1,
    "feed_forward_length": 10240,
    "file_size_bytes": 4714691264,
    "file_size_mb": 4496.28,
    "finetune": null,
    "format": "gguf",
    "gguf_version": 3,
    "has_chat_template": true,
    "head_count": 8,
    "head_count_kv": 2,
    "header_kv_dir_bytes": 15820160,
    "kv_count": 59,
    "kv_meta": {
      "keys": [
        "gemma4.attention.head_count",
        "gemma4.attention.head_count_kv",
        "gemma4.attention.key_length",
        "gemma4.attention.key_length_swa",
        "gemma4.attention.layer_norm_rms_epsilon",
        "gemma4.attention.shared_kv_layers",
        "gemma4.attention.sliding_window",
        "gemma4.attention.sliding_window_pattern",
        "gemma4.attention.value_length",
        "gemma4.attention.value_length_swa",
        "gemma4.block_count",
        "gemma4.context_length",
        "gemma4.embedding_length",
        "gemma4.embedding_length_per_layer_input",
        "gemma4.feed_forward_length",
        "gemma4.final_logit_softcapping",
        "gemma4.rope.dimension_count",
        "gemma4.rope.dimension_count_swa",
        "gemma4.rope.freq_base",
        "gemma4.rope.freq_base_swa",
        "general.architecture",
        "general.base_model.0.name",
        "general.base_model.0.organization",
        "general.base_model.0.repo_url",
        "general.base_model.count",
        "general.file_type",
        "general.license",
        "general.name",
        "general.quantization_version",
        "general.sampling.temp",
        "general.sampling.top_k",
        "general.sampling.top_p",
        "general.size_label",
        "general.source.url",
        "general.tags",
        "general.type",
        "general.url",
        "mradermacher.convert_type",
        "mradermacher.quantize_version",
        "mradermacher.quantized_at",
        "mradermacher.quantized_by",
        "mradermacher.quantized_on",
        "quantize.imatrix.chunks_count",
        "quantize.imatrix.dataset",
        "quantize.imatrix.entries_count",
        "quantize.imatrix.file",
        "tokenizer.chat_template",
        "tokenizer.ggml.add_bos_token",
        "tokenizer.ggml.add_space_prefix",
        "tokenizer.ggml.bos_token_id",
        "tokenizer.ggml.eos_token_id",
        "tokenizer.ggml.mask_token_id",
        "tokenizer.ggml.merges",
        "tokenizer.ggml.model",
        "tokenizer.ggml.padding_token_id",
        "tokenizer.ggml.scores",
        "tokenizer.ggml.token_type",
        "tokenizer.ggml.tokens",
        "tokenizer.ggml.unknown_token_id"
      ],
      "kv_count": 59,
      "values_redacted": true
    },
    "license": "apache-2.0",
    "max_tensor_offset": 4698589504,
    "misaligned_tensors": 0,
    "model_name": "Gemma 4 E4B It SDFT_Heretic_RP",
    "offsets_monotone": true,
    "oob_tensor_count": 0,
    "overlapping_tensor_pairs": 0,
    "param_estimate": 7518069290,
    "params_estimate": 7518069290,
    "quantization": "Q6_K",
    "scanner_model_info": {
      "alignment": 32,
      "architecture": "gemma4",
      "basename": null,
      "block_count": 42,
      "bos_token_id": 2,
      "chat_template": "{%- macro format_parameters(properties, required) -%}\n    {%- set standard_keys = [\u0027description\u0027, \u0027type\u0027, \u0027properties\u0027, \u0027required\u0027, \u0027nullable\u0027] -%}\n    {%- set ns = namespace(found_first=false) -%}\n    {%- for key, value in properties | dictsort -%}\n        {%- set add_comma = false -%}\n        {%- if key not in standard_keys -%}\n            {%- if ns.found_first %},{% endif -%}\n            {%- set ns.found_first = true -%}\n            {{ key }}:{\n            {%- if value[\u0027description\u0027] -%}\n                description:\u003c|\"|\u003e{{ value[\u0027description\u0027] }}\u003c|\"|\u003e\n                {%- set add_comma = true -%}\n            {%- endif -%}\n            {%- if value[\u0027nullable\u0027] %}\n                {%- if add_comma %},{%- else -%} {%- set add_comma = true -%} {% endif -%}\n                nullable:true\n            {%- endif -%}\n            {%- if value[\u0027type\u0027] | upper == \u0027STRING\u0027 -%}\n                {%- if value[\u0027enum\u0027] -%}\n                    {%- if add_comma %},{%- else -%} {%- set add_comma = true -%} {% endif -%}\n                    enum:{{ format_argument(value[\u0027enum\u0027]) }}\n                {%- endif -%}\n            {%- elif value[\u0027type\u0027] | upper == \u0027OBJECT\u0027 -%}\n                ,properties:{\n                {%- if value[\u0027properties\u0027] is defined and value[\u0027properties\u0027] is mapping -%}\n                    {{- format_parameters(value[\u0027properties\u0027], value[\u0027required\u0027] | default([])) -}}\n                {%- elif value is mapping -%}\n                    {{- format_parameters(value, value[\u0027required\u0027] | default([])) -}}\n                {%- endif -%}\n                }\n                {%- if value[\u0027required\u0027] -%}\n                    ,required:[\n                    {%- for item in value[\u0027required\u0027] | default([]) -%}\n                        \u003c|\"|\u003e{{- item -}}\u003c|\"|\u003e\n                        {%- if not loop.last %},{% endif -%}\n                    {%- endfor -%}\n                    ]\n                {%- endif -%}\n            {%- elif value[\u0027type\u0027] | upper == \u0027ARRAY\u0027 -%}\n                {%- if value[\u0027items\u0027] is mapping and value[\u0027items\u0027] -%}\n                    ,items:{\n                    {%- set ns_items = namespace(found_first=false) -%}\n                    {%- for item_key, item_value in value[\u0027items\u0027] | dictsort -%}\n                        {%- if item_value is not none -%}\n                            {%- if ns_items.found_first %},{% endif -%}\n                            {%- set ns_items.found_first = true -%}\n                            {%- if item_key == \u0027properties\u0027 -%}\n                                properties:{\n                                {%- if item_value is mapping -%}\n                                    {{- format_parameters(item_value, value[\u0027items\u0027][\u0027required\u0027] | default([])) -}}\n                                {%- endif -%}\n                                }\n                            {%- elif item_key == \u0027required\u0027 -%}\n                                required:[\n                                {%- for req_item in item_value -%}\n                                    \u003c|\"|\u003e{{- req_item -}}\u003c|\"|\u003e\n                                    {%- if not loop.last %},{% endif -%}\n                                {%- endfor -%}\n                                ]\n                            {%- elif item_key == \u0027type\u0027 -%}\n                                {%- if item_value is string -%}\n                                    type:{{ format_argument(item_value | upper) }}\n                                {%- else -%}\n                                    type:{{ format_argument(item_value | map(\u0027upper\u0027) | list) }}\n                                {%- endif -%}\n                            {%- else -%}\n                                {{ item_key }}:{{ format_argument(item_value) }}\n                            {%- endif -%}\n                        {%- endif -%}\n                    {%- endfor -%}\n                    }\n                {%- endif -%}\n            {%- endif -%}\n            {%- if add_comma %},{%- else -%} {%- set add_comma = true -%} {% endif -%}\n            type:\u003c|\"|\u003e{{ value[\u0027type\u0027] | upper }}\u003c|\"|\u003e}\n        {%- endif -%}\n    {%- endfor -%}\n{%- endmacro -%}\n{%- macro format_function_declaration(tool_data) -%}\n    declaration:{{- tool_data[\u0027function\u0027][\u0027name\u0027] -}}{description:\u003c|\"|\u003e{{- tool_data[\u0027function\u0027][\u0027description\u0027] -}}\u003c|\"|\u003e\n    {%- set params = tool_data[\u0027function\u0027][\u0027parameters\u0027] -%}\n    {%- if params -%}\n        ,parameters:{\n        {%- if params[\u0027properties\u0027] -%}\n            properties:{ {{- format_parameters(params[\u0027properties\u0027], params[\u0027required\u0027]) -}} },\n        {%- endif -%}\n        {%- if params[\u0027required\u0027] -%}\n            required:[\n            {%- for item in params[\u0027required\u0027] -%}\n                \u003c|\"|\u003e{{- item -}}\u003c|\"|\u003e\n                {{- \u0027,\u0027 if not loop.last -}}\n            {%- endfor -%}\n            ],\n        {%- endif -%}\n        {%- if params[\u0027type\u0027] -%}\n            type:\u003c|\"|\u003e{{- params[\u0027type\u0027] | upper -}}\u003c|\"|\u003e}\n        {%- endif -%}\n    {%- endif -%}\n    {%- if \u0027response\u0027 in tool_data[\u0027function\u0027] -%}\n        {%- set response_declaration = tool_data[\u0027function\u0027][\u0027response\u0027] -%}\n        ,response:{\n        {%- if response_declaration[\u0027description\u0027] -%}\n            description:\u003c|\"|\u003e{{- response_declaration[\u0027description\u0027] -}}\u003c|\"|\u003e,\n        {%- endif -%}\n        {%- if response_declaration[\u0027type\u0027] | upper == \u0027OBJECT\u0027 -%}\n            type:\u003c|\"|\u003e{{- response_declaration[\u0027type\u0027] | upper -}}\u003c|\"|\u003e}\n        {%- endif -%}\n    {%- endif -%}\n    }\n{%- endmacro -%}\n{%- macro format_argument(argument, escape_keys=True) -%}\n    {%- if argument is string -%}\n        {{- \u0027\u003c|\"|\u003e\u0027 + argument + \u0027\u003c|\"|\u003e\u0027 -}}\n    {%- elif argument is boolean -%}\n        {{- \u0027true\u0027 if argument else \u0027false\u0027 -}}\n    {%- elif argument is mapping -%}\n        {{- \u0027{\u0027 -}}\n        {%- set ns = namespace(found_first=false) -%}\n        {%- for key, value in argument | dictsort -%}\n            {%- if ns.found_first %},{% endif -%}\n            {%- set ns.found_first = true -%}\n            {%- if escape_keys -%}\n                {{- \u0027\u003c|\"|\u003e\u0027 + key + \u0027\u003c|\"|\u003e\u0027 -}}\n            {%- else -%}\n                {{- key -}}\n            {%- endif -%}\n            :{{- format_argument(value, escape_keys=escape_keys) -}}\n        {%- endfor -%}\n        {{- \u0027}\u0027 -}}\n    {%- elif argument is sequence -%}\n        {{- \u0027[\u0027 -}}\n        {%- for item in argument -%}\n            {{- format_argument(item, escape_keys=escape_keys) -}}\n            {%- if not loop.last %},{% endif -%}\n        {%- endfor -%}\n        {{- \u0027]\u0027 -}}\n    {%- else -%}\n        {{- argument -}}\n    {%- endif -%}\n{%- endmacro -%}\n{%- macro strip_thinking(text) -%}\n    {%- set ns = namespace(result=\u0027\u0027) -%}\n    {%- for part in text.split(\u0027\u003cchannel|\u003e\u0027) -%}\n        {%- if \u0027\u003c|channel\u003e\u0027 in part -%}\n            {%- set ns.result = ns.result + part.split(\u0027\u003c|channel\u003e\u0027)[0] -%}\n        {%- else -%}\n            {%- set ns.result = ns.result + part -%}\n        {%- endif -%}\n    {%- endfor -%}\n    {{- ns.result | trim -}}\n{%- endmacro -%}\n\n{%- set ns = namespace(prev_message_type=None) -%}\n{%- set loop_messages = messages -%}\n{{ bos_token }}\n{#- Handle System/Tool Definitions Block -#}\n{%- if (enable_thinking is defined and enable_thinking) or tools or messages[0][\u0027role\u0027] in [\u0027system\u0027, \u0027developer\u0027] -%}\n    {{- \u0027\u003c|turn\u003esystem\\n\u0027 -}}\n\n    {#- Inject Thinking token at the very top of the FIRST system turn -#}\n    {%- if enable_thinking is defined and enable_thinking -%}\n        {{- \u0027\u003c|think|\u003e\u0027 -}}\n        {%- set ns.prev_message_type = \u0027think\u0027 -%}\n    {%- endif -%}\n\n    {%- if messages[0][\u0027role\u0027] in [\u0027system\u0027, \u0027developer\u0027] -%}\n        {{- messages[0][\u0027content\u0027] | trim -}}\n        {%- set loop_messages = messages[1:] -%}\n    {%- endif -%}\n\n    {%- if tools -%}\n        {%- for tool in tools %}\n            {{- \u0027\u003c|tool\u003e\u0027 -}}\n            {{- format_function_declaration(tool) | trim -}}\n            {{- \u0027\u003ctool|\u003e\u0027 -}}\n        {%- endfor %}\n        {%- set ns.prev_message_type = \u0027tool\u0027 -%}\n    {%- endif -%}\n\n    {{- \u0027\u003cturn|\u003e\\n\u0027 -}}\n{%- endif %}\n\n{#- Loop through messages -#}\n{%- for message in loop_messages -%}\n    {%- set ns.prev_message_type = None -%}\n    {%- set role = \u0027model\u0027 if message[\u0027role\u0027] == \u0027assistant\u0027 else message[\u0027role\u0027] -%}\n        {{- \u0027\u003c|turn\u003e\u0027 + role + \u0027\\n\u0027 }}\n\n            {%- if message[\u0027tool_calls\u0027] -%}\n                {%- for tool_call in message[\u0027tool_calls\u0027] -%}\n                    {%- set function = tool_call[\u0027function\u0027] -%}\n                    {{- \u0027\u003c|tool_call\u003ecall:\u0027 + function[\u0027name\u0027] + \u0027{\u0027 -}}\n                    {%- if function[\u0027arguments\u0027] is mapping -%}\n                        {%- set ns_args = namespace(found_first=false) -%}\n                        {%- for key, value in function[\u0027arguments\u0027] | dictsort -%}\n                            {%- if ns_args.found_first %},{% endif -%}\n                            {%- set ns_args.found_first = true -%}\n                            {{- key -}}:{{- format_argument(value, escape_keys=False) -}}\n                        {%- endfor -%}\n                    {%- elif function[\u0027arguments\u0027] is string -%}\n                        {{- function[\u0027arguments\u0027] -}}\n                    {%- endif -%}\n                    {{- \u0027}\u003ctool_call|\u003e\u0027 -}}\n                {%- endfor -%}\n                {%- set ns.prev_message_type = \u0027tool_call\u0027 -%}\n            {%- endif -%}\n\n            {%- if message[\u0027tool_responses\u0027] -%}\n                {#- Tool Response handling -#}\n                {%- for tool_response in message[\u0027tool_responses\u0027] -%}\n                    {{- \u0027\u003c|tool_response\u003e\u0027 -}}\n                    {%- if tool_response[\u0027response\u0027] is mapping -%}\n                        {{- \u0027response:\u0027 + tool_response[\u0027name\u0027] | default(\u0027unknown\u0027) + \u0027{\u0027 -}}\n                        {%- for key, value in tool_response[\u0027response\u0027] | dictsort -%}\n                            {{- key -}}:{{- format_argument(value, escape_keys=False) -}}\n                            {%- if not loop.last %},{% endif -%}\n                        {%- endfor -%}\n                        {{- \u0027}\u0027 -}}\n                    {%- else -%}\n                        {{- \u0027response:\u0027 + tool_response[\u0027name\u0027] | default(\u0027unknown\u0027) + \u0027{value:\u0027 + format_argument(tool_response[\u0027response\u0027], escape_keys=False) + \u0027}\u0027 -}}\n                    {%- endif -%}\n                    {{- \u0027\u003ctool_response|\u003e\u0027 -}}\n                {%- endfor -%}\n                {%- set ns.prev_message_type = \u0027tool_response\u0027 -%}\n            {%- endif -%}\n\n            {%- if message[\u0027content\u0027] is string -%}\n                {%- if role == \u0027model\u0027 -%}\n                    {{- strip_thinking(message[\u0027content\u0027]) -}}\n                {%- else -%}\n                    {{- message[\u0027content\u0027] | trim -}}\n                {%- endif -%}\n            {%- elif message[\u0027content\u0027] is sequence -%}\n                {%- for item in message[\u0027content\u0027] -%}\n                    {%- if item[\u0027type\u0027] == \u0027text\u0027 -%}\n                        {%- if role == \u0027model\u0027 -%}\n                            {{- strip_thinking(item[\u0027text\u0027]) -}}\n                        {%- else -%}\n                            {{- item[\u0027text\u0027] | trim -}}\n                        {%- endif -%}\n                    {%- elif item[\u0027type\u0027] == \u0027image\u0027 -%}\n                        {{- \u0027\\n\\n\u003c|image|\u003e\\n\\n\u0027 -}}\n                        {%- set ns.prev_message_type = \u0027image\u0027 -%}\n                    {%- elif item[\u0027type\u0027] == \u0027audio\u0027 -%}\n                        {{- \u0027\u003c|audio|\u003e\u0027 -}}\n                        {%- set ns.prev_message_type = \u0027audio\u0027 -%}\n                    {%- elif item[\u0027type\u0027] == \u0027video\u0027 -%}\n                        {{- \u0027\\n\\n\u003c|video|\u003e\\n\\n\u0027 -}}\n                        {%- set ns.prev_message_type = \u0027video\u0027 -%}\n                    {%- endif -%}\n                {%- endfor -%}\n            {%- endif -%}\n\n        {%- if not (message[\u0027tool_responses\u0027] and not message[\u0027content\u0027]) -%}\n            {{- \u0027\u003cturn|\u003e\\n\u0027 -}}\n        {%- endif -%}\n{%- endfor -%}\n\n{%- if add_generation_prompt -%}\n    {%- if ns.prev_message_type != \u0027tool_response\u0027 -%}\n        {{- \u0027\u003c|turn\u003emodel\\n\u0027 -}}\n    {%- endif -%}\n{%- endif -%}",
      "context_length": 131072,
      "data_section_start": 15820160,
      "dtype_histogram": {
        "F16": 1,
        "F32": 339,
        "IQ3_S": 289,
        "Q4_K": 89,
        "Q6_K": 2
      },
      "dtype_percentage": {
        "F16": 0.1,
        "F32": 47.1,
        "IQ3_S": 40.1,
        "Q4_K": 12.4,
        "Q6_K": 0.3
      },
      "dtype_stats": {
        "F16": {
          "avg_bits_per_elem": 16.0,
          "bytes": 55050240,
          "count": 1,
          "mb": 52.5
        },
        "F32": {
          "avg_bits_per_elem": 32.016628,
          "bytes": 2264384,
          "count": 339,
          "mb": 2.16
        },
        "IQ3_S": {
          "avg_bits_per_elem": 3.4375,
          "bytes": 1524582400,
          "count": 289,
          "mb": 1453.96
        },
        "Q4_K": {
          "avg_bits_per_elem": 4.5,
          "bytes": 254361600,
          "count": 89,
          "mb": 242.58
        },
        "Q6_K": {
          "avg_bits_per_elem": 6.5625,
          "bytes": 2862612480,
          "count": 2,
          "mb": 2730.0
        }
      },
      "embedding_length": 2560,
      "eos_token_id": 1,
      "feed_forward_length": 10240,
      "file_size_bytes": 4714691264,
      "file_size_mb": 4496.28,
      "finetune": null,
      "has_chat_template": true,
      "head_count": 8,
      "head_count_kv": 2,
      "header_end_offset": 15820160,
      "header_kv_dir_bytes": 15820160,
      "kv_count": 59,
      "kv_keys": [
        "gemma4.attention.head_count",
        "gemma4.attention.head_count_kv",
        "gemma4.attention.key_length",
        "gemma4.attention.key_length_swa",
        "gemma4.attention.layer_norm_rms_epsilon",
        "gemma4.attention.shared_kv_layers",
        "gemma4.attention.sliding_window",
        "gemma4.attention.sliding_window_pattern",
        "gemma4.attention.value_length",
        "gemma4.attention.value_length_swa",
        "gemma4.block_count",
        "gemma4.context_length",
        "gemma4.embedding_length",
        "gemma4.embedding_length_per_layer_input",
        "gemma4.feed_forward_length",
        "gemma4.final_logit_softcapping",
        "gemma4.rope.dimension_count",
        "gemma4.rope.dimension_count_swa",
        "gemma4.rope.freq_base",
        "gemma4.rope.freq_base_swa",
        "general.architecture",
        "general.base_model.0.name",
        "general.base_model.0.organization",
        "general.base_model.0.repo_url",
        "general.base_model.count",
        "general.file_type",
        "general.license",
        "general.name",
        "general.quantization_version",
        "general.sampling.temp",
        "general.sampling.top_k",
        "general.sampling.top_p",
        "general.size_label",
        "general.source.url",
        "general.tags",
        "general.type",
        "general.url",
        "mradermacher.convert_type",
        "mradermacher.quantize_version",
        "mradermacher.quantized_at",
        "mradermacher.quantized_by",
        "mradermacher.quantized_on",
        "quantize.imatrix.chunks_count",
        "quantize.imatrix.dataset",
        "quantize.imatrix.entries_count",
        "quantize.imatrix.file",
        "tokenizer.chat_template",
        "tokenizer.ggml.add_bos_token",
        "tokenizer.ggml.add_space_prefix",
        "tokenizer.ggml.bos_token_id",
        "tokenizer.ggml.eos_token_id",
        "tokenizer.ggml.mask_token_id",
        "tokenizer.ggml.merges",
        "tokenizer.ggml.model",
        "tokenizer.ggml.padding_token_id",
        "tokenizer.ggml.scores",
        "tokenizer.ggml.token_type",
        "tokenizer.ggml.tokens",
        "tokenizer.ggml.unknown_token_id"
      ],
      "license": "apache-2.0",
      "magic": "GGUF",
      "max_tensor_offset": 4698589504,
      "misaligned_tensors": 0,
      "model_name": "Gemma 4 E4B It SDFT_Heretic_RP",
      "offsets_monotone": true,
      "oob_tensor_count": 0,
      "overlapping_tensor_pairs": 0,
      "param_estimate": 7518069290,
      "quantization": "Q6_K",
      "quantization_declared": "IQ3_S",
      "sha256": "434ed4a3938649780c23486849e2d62138d681e3628aa9396a1c83b0d5821f49",
      "sha512": "094b9356abfe1d657d62f3df1bfc2ce340b6534e5b7739bf4fc5e81004081ede25c1ab15fd173f0a58086de6d6b88623c204dc493e98ec811bc69346584d243b",
      "size_label": "7.5B",
      "tensor_bytes_sum": 4698871104,
      "tensor_count": 720,
      "tensor_delta_bytes": 15820160,
      "tensor_delta_mb": 15.09,
      "tensor_preview": [
        {
          "bits_per_elem": 32.0,
          "byte_len": 10240,
          "dims": [
            2560
          ],
          "dtype_id": 0,
          "dtype_name": "F32",
          "index": 0,
          "name": "output_norm.weight",
          "offset": 0
        },
        {
          "bits_per_elem": 16.0,
          "byte_len": 55050240,
          "dims": [
            2560,
            10752
          ],
          "dtype_id": 1,
          "dtype_name": "F16",
          "index": 1,
          "name": "per_layer_model_proj.weight",
          "offset": 10240
        },
        {
          "bits_per_elem": 32.0,
          "byte_len": 1024,
          "dims": [
            256
          ],
          "dtype_id": 0,
          "dtype_name": "F32",
          "index": 2,
          "name": "per_layer_proj_norm.weight",
          "offset": 55060480
        },
        {
          "bits_per_elem": 6.5625,
          "byte_len": 2312110080,
          "dims": [
            10752,
            262144
          ],
          "dtype_id": 14,
          "dtype_name": "Q6_K",
          "index": 3,
          "name": "per_layer_token_embd.weight",
          "offset": 55061504
        },
        {
          "bits_per_elem": 32.0,
          "byte_len": 1024,
          "dims": [
            256
          ],
          "dtype_id": 0,
          "dtype_name": "F32",
          "index": 4,
          "name": "rope_freqs.weight",
          "offset": 2367171584
        },
        {
          "bits_per_elem": 6.5625,
          "byte_len": 550502400,
          "dims": [
            2560,
            262144
          ],
          "dtype_id": 14,
          "dtype_name": "Q6_K",
          "index": 5,
          "name": "token_embd.weight",
          "offset": 2367172608
        },
        {
          "bits_per_elem": 3.4375,
          "byte_len": 563200,
          "dims": [
            2560,
            512
          ],
          "dtype_id": 21,
          "dtype_name": "IQ3_S",
          "index": 6,
          "name": "blk.0.attn_k.weight",
          "offset": 2917675008
        },
        {
          "bits_per_elem": 32.0,
          "byte_len": 1024,
          "dims": [
            256
          ],
          "dtype_id": 0,
          "dtype_name": "F32",
          "index": 7,
          "name": "blk.0.attn_k_norm.weight",
          "offset": 2918238208
        },
        {
          "bits_per_elem": 32.0,
          "byte_len": 10240,
          "dims": [
            2560
          ],
          "dtype_id": 0,
          "dtype_name": "F32",
          "index": 8,
          "name": "blk.0.attn_norm.weight",
          "offset": 2918239232
        },
        {
          "bits_per_elem": 4.5,
          "byte_len": 2949120,
          "dims": [
            2048,
            2560
          ],
          "dtype_id": 12,
          "dtype_name": "Q4_K",
          "index": 9,
          "name": "blk.0.attn_output.weight",
          "offset": 2918249472
        },
        {
          "bits_per_elem": 3.4375,
          "byte_len": 2252800,
          "dims": [
            2560,
            2048
          ],
          "dtype_id": 21,
          "dtype_name": "IQ3_S",
          "index": 10,
          "name": "blk.0.attn_q.weight",
          "offset": 2921198592
        },
        {
          "bits_per_elem": 32.0,
          "byte_len": 1024,
          "dims": [
            256
          ],
          "dtype_id": 0,
          "dtype_name": "F32",
          "index": 11,
          "name": "blk.0.attn_q_norm.weight",
          "offset": 2923451392
        },
        {
          "bits_per_elem": 4.5,
          "byte_len": 737280,
          "dims": [
            2560,
            512
          ],
          "dtype_id": 12,
          "dtype_name": "Q4_K",
          "index": 12,
          "name": "blk.0.attn_v.weight",
          "offset": 2923452416
        },
        {
          "bits_per_elem": 4.5,
          "byte_len": 14745600,
          "dims": [
            10240,
            2560
          ],
          "dtype_id": 12,
          "dtype_name": "Q4_K",
          "index": 13,
          "name": "blk.0.ffn_down.weight",
          "offset": 2924189696
        },
        {
          "bits_per_elem": 3.4375,
          "byte_len": 11264000,
          "dims": [
            2560,
            10240
          ],
          "dtype_id": 21,
          "dtype_name": "IQ3_S",
          "index": 14,
          "name": "blk.0.ffn_gate.weight",
          "offset": 2938935296
        },
        {
          "bits_per_elem": 32.0,
          "byte_len": 10240,
          "dims": [
            2560
          ],
          "dtype_id": 0,
          "dtype_name": "F32",
          "index": 15,
          "name": "blk.0.ffn_norm.weight",
          "offset": 2950199296
        },
        {
          "bits_per_elem": 3.4375,
          "byte_len": 11264000,
          "dims": [
            2560,
            10240
          ],
          "dtype_id": 21,
          "dtype_name": "IQ3_S",
          "index": 16,
          "name": "blk.0.ffn_up.weight",
          "offset": 2950209536
        },
        {
          "bits_per_elem": 3.4375,
          "byte_len": 281600,
          "dims": [
            2560,
            256
          ],
          "dtype_id": 21,
          "dtype_name": "IQ3_S",
          "index": 17,
          "name": "blk.0.inp_gate.weight",
          "offset": 2961473536
        },
        {
          "bits_per_elem": 256.0,
          "byte_len": 32,
          "dims": [
            1
          ],
          "dtype_id": 0,
          "dtype_name": "F32",
          "index": 18,
          "name": "blk.0.layer_output_scale.weight",
          "offset": 2961755136
        },
        {
          "bits_per_elem": 32.0,
          "byte_len": 10240,
          "dims": [
            2560
          ],
          "dtype_id": 0,
          "dtype_name": "F32",
          "index": 19,
          "name": "blk.0.post_attention_norm.weight",
          "offset": 2961755168
        },
        {
          "bits_per_elem": 32.0,
          "byte_len": 10240,
          "dims": [
            2560
          ],
          "dtype_id": 0,
          "dtype_name": "F32",
          "index": 20,
          "name": "blk.0.post_ffw_norm.weight",
          "offset": 2961765408
        },
        {
          "bits_per_elem": 32.0,
          "byte_len": 10240,
          "dims": [
            2560
          ],
          "dtype_id": 0,
          "dtype_name": "F32",
          "index": 21,
          "name": "blk.0.post_norm.weight",
          "offset": 2961775648
        },
        {
          "bits_per_elem": 3.4375,
          "byte_len": 281600,
          "dims": [
            256,
            2560
          ],
          "dtype_id": 21,
          "dtype_name": "IQ3_S",
          "index": 22,
          "name": "blk.0.proj.weight",
          "offset": 2961785888
        },
        {
          "bits_per_elem": 3.4375,
          "byte_len": 563200,
          "dims": [
            2560,
            512
          ],
          "dtype_id": 21,
          "dtype_name": "IQ3_S",
          "index": 23,
          "name": "blk.1.attn_k.weight",
          "offset": 2962067488
        },
        {
          "bits_per_elem": 32.0,
          "byte_len": 1024,
          "dims": [
            256
          ],
          "dtype_id": 0,
          "dtype_name": "F32",
          "index": 24,
          "name": "blk.1.attn_k_norm.weight",
          "offset": 2962630688
        },
        {
          "bits_per_elem": 32.0,
          "byte_len": 10240,
          "dims": [
            2560
          ],
          "dtype_id": 0,
          "dtype_name": "F32",
          "index": 25,
          "name": "blk.1.attn_norm.weight",
          "offset": 2962631712
        },
        {
          "bits_per_elem": 4.5,
          "byte_len": 2949120,
          "dims": [
            2048,
            2560
          ],
          "dtype_id": 12,
          "dtype_name": "Q4_K",
          "index": 26,
          "name": "blk.1.attn_output.weight",
          "offset": 2962641952
        },
        {
          "bits_per_elem": 3.4375,
          "byte_len": 2252800,
          "dims": [
            2560,
            2048
          ],
          "dtype_id": 21,
          "dtype_name": "IQ3_S",
          "index": 27,
          "name": "blk.1.attn_q.weight",
          "offset": 2965591072
        },
        {
          "bits_per_elem": 32.0,
          "byte_len": 1024,
          "dims": [
            256
          ],
          "dtype_id": 0,
          "dtype_name": "F32",
          "index": 28,
          "name": "blk.1.attn_q_norm.weight",
          "offset": 2967843872
        },
        {
          "bits_per_elem": 4.5,
          "byte_len": 737280,
          "dims": [
            2560,
            512
          ],
          "dtype_id": 12,
          "dtype_name": "Q4_K",
          "index": 29,
          "name": "blk.1.attn_v.weight",
          "offset": 2967844896
        },
        {
          "bits_per_elem": 4.5,
          "byte_len": 14745600,
          "dims": [
            10240,
            2560
          ],
          "dtype_id": 12,
          "dtype_name": "Q4_K",
          "index": 30,
          "name": "blk.1.ffn_down.weight",
          "offset": 2968582176
        },
        {
          "bits_per_elem": 3.4375,
          "byte_len": 11264000,
          "dims": [
            2560,
            10240
          ],
          "dtype_id": 21,
          "dtype_name": "IQ3_S",
          "index": 31,
          "name": "blk.1.ffn_gate.weight",
          "offset": 2983327776
        },
        {
          "bits_per_elem": 32.0,
          "byte_len": 10240,
          "dims": [
            2560
          ],
          "dtype_id": 0,
          "dtype_name": "F32",
          "index": 32,
          "name": "blk.1.ffn_norm.weight",
          "offset": 2994591776
        },
        {
          "bits_per_elem": 3.4375,
          "byte_len": 11264000,
          "dims": [
            2560,
            10240
          ],
          "dtype_id": 21,
          "dtype_name": "IQ3_S",
          "index": 33,
          "name": "blk.1.ffn_up.weight",
          "offset": 2994602016
        },
        {
          "bits_per_elem": 3.4375,
          "byte_len": 281600,
          "dims": [
            2560,
            256
          ],
          "dtype_id": 21,
          "dtype_name": "IQ3_S",
          "index": 34,
          "name": "blk.1.inp_gate.weight",
          "offset": 3005866016
        },
        {
          "bits_per_elem": 256.0,
          "byte_len": 32,
          "dims": [
            1
          ],
          "dtype_id": 0,
          "dtype_name": "F32",
          "index": 35,
          "name": "blk.1.layer_output_scale.weight",
          "offset": 3006147616
        },
        {
          "bits_per_elem": 32.0,
          "byte_len": 10240,
          "dims": [
            2560
          ],
          "dtype_id": 0,
          "dtype_name": "F32",
          "index": 36,
          "name": "blk.1.post_attention_norm.weight",
          "offset": 3006147648
        },
        {
          "bits_per_elem": 32.0,
          "byte_len": 10240,
          "dims": [
            2560
          ],
          "dtype_id": 0,
          "dtype_name": "F32",
          "index": 37,
          "name": "blk.1.post_ffw_norm.weight",
          "offset": 3006157888
        },
        {
          "bits_per_elem": 32.0,
          "byte_len": 10240,
          "dims": [
            2560
          ],
          "dtype_id": 0,
          "dtype_name": "F32",
          "index": 38,
          "name": "blk.1.post_norm.weight",
          "offset": 3006168128
        },
        {
          "bits_per_elem": 3.4375,
          "byte_len": 281600,
          "dims": [
            256,
            2560
          ],
          "dtype_id": 21,
          "dtype_name": "IQ3_S",
          "index": 39,
          "name": "blk.1.proj.weight",
          "offset": 3006178368
        },
        {
          "bits_per_elem": 3.4375,
          "byte_len": 563200,
          "dims": [
            2560,
            512
          ],
          "dtype_id": 21,
          "dtype_name": "IQ3_S",
          "index": 40,
          "name": "blk.2.attn_k.weight",
          "offset": 3006459968
        },
        {
          "bits_per_elem": 32.0,
          "byte_len": 1024,
          "dims": [
            256
          ],
          "dtype_id": 0,
          "dtype_name": "F32",
          "index": 41,
          "name": "blk.2.attn_k_norm.weight",
          "offset": 3007023168
        },
        {
          "bits_per_elem": 32.0,
          "byte_len": 10240,
          "dims": [
            2560
          ],
          "dtype_id": 0,
          "dtype_name": "F32",
          "index": 42,
          "name": "blk.2.attn_norm.weight",
          "offset": 3007024192
        },
        {
          "bits_per_elem": 4.5,
          "byte_len": 2949120,
          "dims": [
            2048,
            2560
          ],
          "dtype_id": 12,
          "dtype_name": "Q4_K",
          "index": 43,
          "name": "blk.2.attn_output.weight",
          "offset": 3007034432
        },
        {
          "bits_per_elem": 3.4375,
          "byte_len": 2252800,
          "dims": [
            2560,
            2048
          ],
          "dtype_id": 21,
          "dtype_name": "IQ3_S",
          "index": 44,
          "name": "blk.2.attn_q.weight",
          "offset": 3009983552
        },
        {
          "bits_per_elem": 32.0,
          "byte_len": 1024,
          "dims": [
            256
          ],
          "dtype_id": 0,
          "dtype_name": "F32",
          "index": 45,
          "name": "blk.2.attn_q_norm.weight",
          "offset": 3012236352
        },
        {
          "bits_per_elem": 4.5,
          "byte_len": 737280,
          "dims": [
            2560,
            512
          ],
          "dtype_id": 12,
          "dtype_name": "Q4_K",
          "index": 46,
          "name": "blk.2.attn_v.weight",
          "offset": 3012237376
        },
        {
          "bits_per_elem": 4.5,
          "byte_len": 14745600,
          "dims": [
            10240,
            2560
          ],
          "dtype_id": 12,
          "dtype_name": "Q4_K",
          "index": 47,
          "name": "blk.2.ffn_down.weight",
          "offset": 3012974656
        },
        {
          "bits_per_elem": 3.4375,
          "byte_len": 11264000,
          "dims": [
            2560,
            10240
          ],
          "dtype_id": 21,
          "dtype_name": "IQ3_S",
          "index": 48,
          "name": "blk.2.ffn_gate.weight",
          "offset": 3027720256
        },
        {
          "bits_per_elem": 32.0,
          "byte_len": 10240,
          "dims": [
            2560
          ],
          "dtype_id": 0,
          "dtype_name": "F32",
          "index": 49,
          "name": "blk.2.ffn_norm.weight",
          "offset": 3038984256
        }
      ],
      "tokenizer_model": "gemma4",
      "top_tensors_by_size": [
        {
          "byte_len": 2312110080,
          "dims": [
            10752,
            262144
          ],
          "dtype": "Q6_K",
          "n_elem": 2818572288,
          "name": "per_layer_token_embd.weight"
        },
        {
          "byte_len": 550502400,
          "dims": [
            2560,
            262144
          ],
          "dtype": "Q6_K",
          "n_elem": 671088640,
          "name": "token_embd.weight"
        },
        {
          "byte_len": 55050240,
          "dims": [
            2560,
            10752
          ],
          "dtype": "F16",
          "n_elem": 27525120,
          "name": "per_layer_model_proj.weight"
        },
        {
          "byte_len": 14745600,
          "dims": [
            10240,
            2560
          ],
          "dtype": "Q4_K",
          "n_elem": 26214400,
          "name": "blk.0.ffn_down.weight"
        },
        {
          "byte_len": 14745600,
          "dims": [
            10240,
            2560
          ],
          "dtype": "Q4_K",
          "n_elem": 26214400,
          "name": "blk.1.ffn_down.weight"
        },
        {
          "byte_len": 14745600,
          "dims": [
            10240,
            2560
          ],
          "dtype": "Q4_K",
          "n_elem": 26214400,
          "name": "blk.2.ffn_down.weight"
        },
        {
          "byte_len": 14745600,
          "dims": [
            10240,
            2560
          ],
          "dtype": "Q4_K",
          "n_elem": 26214400,
          "name": "blk.3.ffn_down.weight"
        },
        {
          "byte_len": 14745600,
          "dims": [
            10240,
            2560
          ],
          "dtype": "Q4_K",
          "n_elem": 26214400,
          "name": "blk.4.ffn_down.weight"
        },
        {
          "byte_len": 11264000,
          "dims": [
            2560,
            10240
          ],
          "dtype": "IQ3_S",
          "n_elem": 26214400,
          "name": "blk.0.ffn_gate.weight"
        },
        {
          "byte_len": 11264000,
          "dims": [
            2560,
            10240
          ],
          "dtype": "IQ3_S",
          "n_elem": 26214400,
          "name": "blk.0.ffn_up.weight"
        }
      ],
      "trailing_bytes": 0,
      "unaccounted_bytes": 0,
      "version": 3,
      "vocab_size": 262144
    },
    "sha256": "434ed4a3938649780c23486849e2d62138d681e3628aa9396a1c83b0d5821f49",
    "sha512": "094b9356abfe1d657d62f3df1bfc2ce340b6534e5b7739bf4fc5e81004081ede25c1ab15fd173f0a58086de6d6b88623c204dc493e98ec811bc69346584d243b",
    "size_label": "7.5B",
    "tensor_bytes_sum": 4698871104,
    "tensor_count": 720,
    "tensor_delta_bytes": 15820160,
    "tensor_delta_mb": 15.09,
    "tensor_preview": [
      {
        "bits_per_elem": 32.0,
        "byte_len": 10240,
        "dims": [
          2560
        ],
        "dtype_id": 0,
        "dtype_name": "F32",
        "index": 0,
        "name": "output_norm.weight",
        "offset": 0
      },
      {
        "bits_per_elem": 16.0,
        "byte_len": 55050240,
        "dims": [
          2560,
          10752
        ],
        "dtype_id": 1,
        "dtype_name": "F16",
        "index": 1,
        "name": "per_layer_model_proj.weight",
        "offset": 10240
      },
      {
        "bits_per_elem": 32.0,
        "byte_len": 1024,
        "dims": [
          256
        ],
        "dtype_id": 0,
        "dtype_name": "F32",
        "index": 2,
        "name": "per_layer_proj_norm.weight",
        "offset": 55060480
      },
      {
        "bits_per_elem": 6.5625,
        "byte_len": 2312110080,
        "dims": [
          10752,
          262144
        ],
        "dtype_id": 14,
        "dtype_name": "Q6_K",
        "index": 3,
        "name": "per_layer_token_embd.weight",
        "offset": 55061504
      },
      {
        "bits_per_elem": 32.0,
        "byte_len": 1024,
        "dims": [
          256
        ],
        "dtype_id": 0,
        "dtype_name": "F32",
        "index": 4,
        "name": "rope_freqs.weight",
        "offset": 2367171584
      },
      {
        "bits_per_elem": 6.5625,
        "byte_len": 550502400,
        "dims": [
          2560,
          262144
        ],
        "dtype_id": 14,
        "dtype_name": "Q6_K",
        "index": 5,
        "name": "token_embd.weight",
        "offset": 2367172608
      },
      {
        "bits_per_elem": 3.4375,
        "byte_len": 563200,
        "dims": [
          2560,
          512
        ],
        "dtype_id": 21,
        "dtype_name": "IQ3_S",
        "index": 6,
        "name": "blk.0.attn_k.weight",
        "offset": 2917675008
      },
      {
        "bits_per_elem": 32.0,
        "byte_len": 1024,
        "dims": [
          256
        ],
        "dtype_id": 0,
        "dtype_name": "F32",
        "index": 7,
        "name": "blk.0.attn_k_norm.weight",
        "offset": 2918238208
      },
      {
        "bits_per_elem": 32.0,
        "byte_len": 10240,
        "dims": [
          2560
        ],
        "dtype_id": 0,
        "dtype_name": "F32",
        "index": 8,
        "name": "blk.0.attn_norm.weight",
        "offset": 2918239232
      },
      {
        "bits_per_elem": 4.5,
        "byte_len": 2949120,
        "dims": [
          2048,
          2560
        ],
        "dtype_id": 12,
        "dtype_name": "Q4_K",
        "index": 9,
        "name": "blk.0.attn_output.weight",
        "offset": 2918249472
      },
      {
        "bits_per_elem": 3.4375,
        "byte_len": 2252800,
        "dims": [
          2560,
          2048
        ],
        "dtype_id": 21,
        "dtype_name": "IQ3_S",
        "index": 10,
        "name": "blk.0.attn_q.weight",
        "offset": 2921198592
      },
      {
        "bits_per_elem": 32.0,
        "byte_len": 1024,
        "dims": [
          256
        ],
        "dtype_id": 0,
        "dtype_name": "F32",
        "index": 11,
        "name": "blk.0.attn_q_norm.weight",
        "offset": 2923451392
      },
      {
        "bits_per_elem": 4.5,
        "byte_len": 737280,
        "dims": [
          2560,
          512
        ],
        "dtype_id": 12,
        "dtype_name": "Q4_K",
        "index": 12,
        "name": "blk.0.attn_v.weight",
        "offset": 2923452416
      },
      {
        "bits_per_elem": 4.5,
        "byte_len": 14745600,
        "dims": [
          10240,
          2560
        ],
        "dtype_id": 12,
        "dtype_name": "Q4_K",
        "index": 13,
        "name": "blk.0.ffn_down.weight",
        "offset": 2924189696
      },
      {
        "bits_per_elem": 3.4375,
        "byte_len": 11264000,
        "dims": [
          2560,
          10240
        ],
        "dtype_id": 21,
        "dtype_name": "IQ3_S",
        "index": 14,
        "name": "blk.0.ffn_gate.weight",
        "offset": 2938935296
      },
      {
        "bits_per_elem": 32.0,
        "byte_len": 10240,
        "dims": [
          2560
        ],
        "dtype_id": 0,
        "dtype_name": "F32",
        "index": 15,
        "name": "blk.0.ffn_norm.weight",
        "offset": 2950199296
      },
      {
        "bits_per_elem": 3.4375,
        "byte_len": 11264000,
        "dims": [
          2560,
          10240
        ],
        "dtype_id": 21,
        "dtype_name": "IQ3_S",
        "index": 16,
        "name": "blk.0.ffn_up.weight",
        "offset": 2950209536
      },
      {
        "bits_per_elem": 3.4375,
        "byte_len": 281600,
        "dims": [
          2560,
          256
        ],
        "dtype_id": 21,
        "dtype_name": "IQ3_S",
        "index": 17,
        "name": "blk.0.inp_gate.weight",
        "offset": 2961473536
      },
      {
        "bits_per_elem": 256.0,
        "byte_len": 32,
        "dims": [
          1
        ],
        "dtype_id": 0,
        "dtype_name": "F32",
        "index": 18,
        "name": "blk.0.layer_output_scale.weight",
        "offset": 2961755136
      },
      {
        "bits_per_elem": 32.0,
        "byte_len": 10240,
        "dims": [
          2560
        ],
        "dtype_id": 0,
        "dtype_name": "F32",
        "index": 19,
        "name": "blk.0.post_attention_norm.weight",
        "offset": 2961755168
      },
      {
        "bits_per_elem": 32.0,
        "byte_len": 10240,
        "dims": [
          2560
        ],
        "dtype_id": 0,
        "dtype_name": "F32",
        "index": 20,
        "name": "blk.0.post_ffw_norm.weight",
        "offset": 2961765408
      },
      {
        "bits_per_elem": 32.0,
        "byte_len": 10240,
        "dims": [
          2560
        ],
        "dtype_id": 0,
        "dtype_name": "F32",
        "index": 21,
        "name": "blk.0.post_norm.weight",
        "offset": 2961775648
      },
      {
        "bits_per_elem": 3.4375,
        "byte_len": 281600,
        "dims": [
          256,
          2560
        ],
        "dtype_id": 21,
        "dtype_name": "IQ3_S",
        "index": 22,
        "name": "blk.0.proj.weight",
        "offset": 2961785888
      },
      {
        "bits_per_elem": 3.4375,
        "byte_len": 563200,
        "dims": [
          2560,
          512
        ],
        "dtype_id": 21,
        "dtype_name": "IQ3_S",
        "index": 23,
        "name": "blk.1.attn_k.weight",
        "offset": 2962067488
      },
      {
        "bits_per_elem": 32.0,
        "byte_len": 1024,
        "dims": [
          256
        ],
        "dtype_id": 0,
        "dtype_name": "F32",
        "index": 24,
        "name": "blk.1.attn_k_norm.weight",
        "offset": 2962630688
      },
      {
        "bits_per_elem": 32.0,
        "byte_len": 10240,
        "dims": [
          2560
        ],
        "dtype_id": 0,
        "dtype_name": "F32",
        "index": 25,
        "name": "blk.1.attn_norm.weight",
        "offset": 2962631712
      },
      {
        "bits_per_elem": 4.5,
        "byte_len": 2949120,
        "dims": [
          2048,
          2560
        ],
        "dtype_id": 12,
        "dtype_name": "Q4_K",
        "index": 26,
        "name": "blk.1.attn_output.weight",
        "offset": 2962641952
      },
      {
        "bits_per_elem": 3.4375,
        "byte_len": 2252800,
        "dims": [
          2560,
          2048
        ],
        "dtype_id": 21,
        "dtype_name": "IQ3_S",
        "index": 27,
        "name": "blk.1.attn_q.weight",
        "offset": 2965591072
      },
      {
        "bits_per_elem": 32.0,
        "byte_len": 1024,
        "dims": [
          256
        ],
        "dtype_id": 0,
        "dtype_name": "F32",
        "index": 28,
        "name": "blk.1.attn_q_norm.weight",
        "offset": 2967843872
      },
      {
        "bits_per_elem": 4.5,
        "byte_len": 737280,
        "dims": [
          2560,
          512
        ],
        "dtype_id": 12,
        "dtype_name": "Q4_K",
        "index": 29,
        "name": "blk.1.attn_v.weight",
        "offset": 2967844896
      },
      {
        "bits_per_elem": 4.5,
        "byte_len": 14745600,
        "dims": [
          10240,
          2560
        ],
        "dtype_id": 12,
        "dtype_name": "Q4_K",
        "index": 30,
        "name": "blk.1.ffn_down.weight",
        "offset": 2968582176
      },
      {
        "bits_per_elem": 3.4375,
        "byte_len": 11264000,
        "dims": [
          2560,
          10240
        ],
        "dtype_id": 21,
        "dtype_name": "IQ3_S",
        "index": 31,
        "name": "blk.1.ffn_gate.weight",
        "offset": 2983327776
      },
      {
        "bits_per_elem": 32.0,
        "byte_len": 10240,
        "dims": [
          2560
        ],
        "dtype_id": 0,
        "dtype_name": "F32",
        "index": 32,
        "name": "blk.1.ffn_norm.weight",
        "offset": 2994591776
      },
      {
        "bits_per_elem": 3.4375,
        "byte_len": 11264000,
        "dims": [
          2560,
          10240
        ],
        "dtype_id": 21,
        "dtype_name": "IQ3_S",
        "index": 33,
        "name": "blk.1.ffn_up.weight",
        "offset": 2994602016
      },
      {
        "bits_per_elem": 3.4375,
        "byte_len": 281600,
        "dims": [
          2560,
          256
        ],
        "dtype_id": 21,
        "dtype_name": "IQ3_S",
        "index": 34,
        "name": "blk.1.inp_gate.weight",
        "offset": 3005866016
      },
      {
        "bits_per_elem": 256.0,
        "byte_len": 32,
        "dims": [
          1
        ],
        "dtype_id": 0,
        "dtype_name": "F32",
        "index": 35,
        "name": "blk.1.layer_output_scale.weight",
        "offset": 3006147616
      },
      {
        "bits_per_elem": 32.0,
        "byte_len": 10240,
        "dims": [
          2560
        ],
        "dtype_id": 0,
        "dtype_name": "F32",
        "index": 36,
        "name": "blk.1.post_attention_norm.weight",
        "offset": 3006147648
      },
      {
        "bits_per_elem": 32.0,
        "byte_len": 10240,
        "dims": [
          2560
        ],
        "dtype_id": 0,
        "dtype_name": "F32",
        "index": 37,
        "name": "blk.1.post_ffw_norm.weight",
        "offset": 3006157888
      },
      {
        "bits_per_elem": 32.0,
        "byte_len": 10240,
        "dims": [
          2560
        ],
        "dtype_id": 0,
        "dtype_name": "F32",
        "index": 38,
        "name": "blk.1.post_norm.weight",
        "offset": 3006168128
      },
      {
        "bits_per_elem": 3.4375,
        "byte_len": 281600,
        "dims": [
          256,
          2560
        ],
        "dtype_id": 21,
        "dtype_name": "IQ3_S",
        "index": 39,
        "name": "blk.1.proj.weight",
        "offset": 3006178368
      },
      {
        "bits_per_elem": 3.4375,
        "byte_len": 563200,
        "dims": [
          2560,
          512
        ],
        "dtype_id": 21,
        "dtype_name": "IQ3_S",
        "index": 40,
        "name": "blk.2.attn_k.weight",
        "offset": 3006459968
      },
      {
        "bits_per_elem": 32.0,
        "byte_len": 1024,
        "dims": [
          256
        ],
        "dtype_id": 0,
        "dtype_name": "F32",
        "index": 41,
        "name": "blk.2.attn_k_norm.weight",
        "offset": 3007023168
      },
      {
        "bits_per_elem": 32.0,
        "byte_len": 10240,
        "dims": [
          2560
        ],
        "dtype_id": 0,
        "dtype_name": "F32",
        "index": 42,
        "name": "blk.2.attn_norm.weight",
        "offset": 3007024192
      },
      {
        "bits_per_elem": 4.5,
        "byte_len": 2949120,
        "dims": [
          2048,
          2560
        ],
        "dtype_id": 12,
        "dtype_name": "Q4_K",
        "index": 43,
        "name": "blk.2.attn_output.weight",
        "offset": 3007034432
      },
      {
        "bits_per_elem": 3.4375,
        "byte_len": 2252800,
        "dims": [
          2560,
          2048
        ],
        "dtype_id": 21,
        "dtype_name": "IQ3_S",
        "index": 44,
        "name": "blk.2.attn_q.weight",
        "offset": 3009983552
      },
      {
        "bits_per_elem": 32.0,
        "byte_len": 1024,
        "dims": [
          256
        ],
        "dtype_id": 0,
        "dtype_name": "F32",
        "index": 45,
        "name": "blk.2.attn_q_norm.weight",
        "offset": 3012236352
      },
      {
        "bits_per_elem": 4.5,
        "byte_len": 737280,
        "dims": [
          2560,
          512
        ],
        "dtype_id": 12,
        "dtype_name": "Q4_K",
        "index": 46,
        "name": "blk.2.attn_v.weight",
        "offset": 3012237376
      },
      {
        "bits_per_elem": 4.5,
        "byte_len": 14745600,
        "dims": [
          10240,
          2560
        ],
        "dtype_id": 12,
        "dtype_name": "Q4_K",
        "index": 47,
        "name": "blk.2.ffn_down.weight",
        "offset": 3012974656
      },
      {
        "bits_per_elem": 3.4375,
        "byte_len": 11264000,
        "dims": [
          2560,
          10240
        ],
        "dtype_id": 21,
        "dtype_name": "IQ3_S",
        "index": 48,
        "name": "blk.2.ffn_gate.weight",
        "offset": 3027720256
      },
      {
        "bits_per_elem": 32.0,
        "byte_len": 10240,
        "dims": [
          2560
        ],
        "dtype_id": 0,
        "dtype_name": "F32",
        "index": 49,
        "name": "blk.2.ffn_norm.weight",
        "offset": 3038984256
      }
    ],
    "tokenizer_model": "gemma4",
    "top_tensors_by_size": [
      {
        "byte_len": 2312110080,
        "dims": [
          10752,
          262144
        ],
        "dtype": "Q6_K",
        "n_elem": 2818572288,
        "name": "per_layer_token_embd.weight"
      },
      {
        "byte_len": 550502400,
        "dims": [
          2560,
          262144
        ],
        "dtype": "Q6_K",
        "n_elem": 671088640,
        "name": "token_embd.weight"
      },
      {
        "byte_len": 55050240,
        "dims": [
          2560,
          10752
        ],
        "dtype": "F16",
        "n_elem": 27525120,
        "name": "per_layer_model_proj.weight"
      },
      {
        "byte_len": 14745600,
        "dims": [
          10240,
          2560
        ],
        "dtype": "Q4_K",
        "n_elem": 26214400,
        "name": "blk.0.ffn_down.weight"
      },
      {
        "byte_len": 14745600,
        "dims": [
          10240,
          2560
        ],
        "dtype": "Q4_K",
        "n_elem": 26214400,
        "name": "blk.1.ffn_down.weight"
      },
      {
        "byte_len": 14745600,
        "dims": [
          10240,
          2560
        ],
        "dtype": "Q4_K",
        "n_elem": 26214400,
        "name": "blk.2.ffn_down.weight"
      },
      {
        "byte_len": 14745600,
        "dims": [
          10240,
          2560
        ],
        "dtype": "Q4_K",
        "n_elem": 26214400,
        "name": "blk.3.ffn_down.weight"
      },
      {
        "byte_len": 14745600,
        "dims": [
          10240,
          2560
        ],
        "dtype": "Q4_K",
        "n_elem": 26214400,
        "name": "blk.4.ffn_down.weight"
      },
      {
        "byte_len": 11264000,
        "dims": [
          2560,
          10240
        ],
        "dtype": "IQ3_S",
        "n_elem": 26214400,
        "name": "blk.0.ffn_gate.weight"
      },
      {
        "byte_len": 11264000,
        "dims": [
          2560,
          10240
        ],
        "dtype": "IQ3_S",
        "n_elem": 26214400,
        "name": "blk.0.ffn_up.weight"
      }
    ],
    "trailing_bytes": 0,
    "unaccounted_bytes": 0,
    "values_redacted": true,
    "vocab_size": 262144
  },
  "report": {
    "bullets": [
      "GGUF v3 model | architecture: gemma4 | size: 7.5B | quantization: Q6_K | 720 tensors | 3 warnings / 5 informational",
      "Size: 4496.28 MB",
      "SHA256: 434ed4a393864978..."
    ],
    "recommendations": [
      "Review template before use in agentic pipelines. Use sandboxed Jinja2 rendering and compare against the upstream author\u0027s reference template.",
      "URLs in model metadata may be used for tracking or exfiltration when metadata is rendered; do not auto-fetch",
      "Inspect outlier tensor names and compare against reference architecture"
    ],
    "summary": "GGUF v3 model | architecture: gemma4 | size: 7.5B | quantization: Q6_K | 720 tensors | 3 warnings / 5 informational"
  },
  "scan_context": {
    "duration_ms": 82480,
    "ended_at": "2026-09-15T05:25:04+00:00Z",
    "limits": {
      "sandbox": true,
      "timeout_ms": 60000
    },
    "mode": "static-deep",
    "started_at": "2026-09-15T05:25:04+00:00Z"
  },
  "security": {
    "findings": [
      {
        "category": "Architecture",
        "detail": "head_count=8, head_count_kv=2 \u2014 requires GQA-aware runtime",
        "recommendation": "Ensure your runtime (llama.cpp \u003e= b1500, Ollama \u003e= 0.1.30) supports GQA",
        "severity": "INFO",
        "title": "Grouped Query Attention (GQA) detected"
      },
      {
        "category": "Metadata Security",
        "detail": "Template uses tool_calls, tool responses, namespace() \u2014 the full pattern for agentic pipelines. These are legitimate HuggingFace constructs but represent the highest-complexity template class. Risk is context-dependent: negligible for static inference, meaningful in autonomous-agent deployments where user-controlled content can reach the template renderer.",
        "recommendation": "Review template before use in agentic pipelines. Use sandboxed Jinja2 rendering and compare against the upstream author\u0027s reference template.",
        "severity": "MEDIUM",
        "title": "Complex agentic chat template \u2014 context-dependent risk"
      },
      {
        "category": "Metadata Security",
        "detail": "Fields: [\u0027general.base_model.0.repo_url\u0027, \u0027general.url\u0027, \u0027general.source.url\u0027]",
        "recommendation": "URLs in model metadata may be used for tracking or exfiltration when metadata is rendered; do not auto-fetch",
        "severity": "MEDIUM",
        "title": "External URLs embedded in 3 KV metadata field(s)"
      },
      {
        "category": "Embedding Size",
        "detail": "vocab=262,144 x emb_dim=10,752 x 0.8203125 B/elem -\u003e expected 2,312,110,080 B \u2014 size is consistent with architecture",
        "recommendation": "No action needed; large embeddings are expected for large-vocabulary models",
        "severity": "INFO",
        "title": "Large embedding tensor \u0027per_layer_token_embd.weight\u0027: 2,312,110,080 bytes"
      },
      {
        "category": "Embedding Size",
        "detail": "vocab=262,144 x emb_dim=2,560 x 0.8203125 B/elem -\u003e expected 550,502,400 B \u2014 size is consistent with architecture",
        "recommendation": "No action needed; large embeddings are expected for large-vocabulary models",
        "severity": "INFO",
        "title": "Large embedding tensor \u0027token_embd.weight\u0027: 550,502,400 bytes"
      },
      {
        "category": "Statistical Anomaly",
        "detail": "Extreme outlier tensors may contain embedded payloads or crafted weight matrices",
        "recommendation": "Inspect outlier tensor names and compare against reference architecture",
        "severity": "LOW",
        "title": "134 weight tensor(s) are \u003e20x the median size: per_layer_model_proj.weight, blk.0.ffn_down.weight, blk.0.ffn_gate.weight"
      },
      {
        "category": "Quantization",
        "detail": "Quantization reduces size and inference cost at the expense of some accuracy",
        "recommendation": "Verify the quantization level is appropriate for your use case (Q4_K_M is a common balanced choice)",
        "severity": "INFO",
        "title": "Model is quantized: IQ3_S, Q4_K, Q6_K"
      },
      {
        "category": "Quantization",
        "detail": "The GGUF general.file_type field declares \u0027IQ3_S\u0027 but the actual tensors in this file are \u0027Q6_K\u0027. This is common for multi-file model packages where the projector/adapter GGUF inherits the main model\u0027s file_type metadata. The effective dtype shown in the report reflects actual tensor storage.",
        "recommendation": "No action required if this is a companion file (mmproj, adapter). For standalone models, this may indicate incorrect GGUF export.",
        "severity": "INFO",
        "title": "Declared file_type (IQ3_S) does not match effective tensor dtypes (Q6_K)"
      }
    ],
    "grade": "B",
    "score": 81
  },
  "source": {
    "file_sha": null,
    "filename": "gemma-4-E4B-it-SDFT_Heretic_RP.i1-IQ3_M.gguf",
    "hashes": {
      "sha256": "434ed4a3938649780c23486849e2d62138d681e3628aa9396a1c83b0d5821f49"
    },
    "hf_provenance": {
      "downloads": 1156,
      "file_sha": null,
      "has_config": false,
      "has_readme": true,
      "has_tokenizer_config": false,
      "likes": 4,
      "pipeline_tag": null,
      "repo_id": "mradermacher/gemma-4-E4B-it-SDFT_Heretic_RP-i1-GGUF",
      "repo_license": "apache-2.0",
      "repo_sha": "a6356be002df20b7c3abaa96b90b0ff328e2b0cf",
      "siblings": [
        {
          "name": ".gitattributes",
          "sha": null,
          "size": 2819
        },
        {
          "name": "README.md",
          "sha": null,
          "size": 5362
        },
        {
          "name": "gemma-4-E4B-it-SDFT_Heretic_RP.i1-IQ3_M.gguf",
          "sha": null,
          "size": 4714691264
        },
        {
          "name": "gemma-4-E4B-it-SDFT_Heretic_RP.i1-IQ3_S.gguf",
          "sha": null,
          "size": 4663163584
        },
        {
          "name": "gemma-4-E4B-it-SDFT_Heretic_RP.i1-IQ4_NL.gguf",
          "sha": null,
          "size": 5193953984
        },
        {
          "name": "gemma-4-E4B-it-SDFT_Heretic_RP.i1-IQ4_XS.gguf",
          "sha": null,
          "size": 5070951104
        },
        {
          "name": "gemma-4-E4B-it-SDFT_Heretic_RP.i1-Q2_K.gguf",
          "sha": null,
          "size": 4401316544
        },
        {
          "name": "gemma-4-E4B-it-SDFT_Heretic_RP.i1-Q3_K_L.gguf",
          "sha": null,
          "size": 5021276864
        },
        {
          "name": "gemma-4-E4B-it-SDFT_Heretic_RP.i1-Q3_K_M.gguf",
          "sha": null,
          "size": 4850391744
        },
        {
          "name": "gemma-4-E4B-it-SDFT_Heretic_RP.i1-Q3_K_S.gguf",
          "sha": null,
          "size": 4654633664
        },
        {
          "name": "gemma-4-E4B-it-SDFT_Heretic_RP.i1-Q4_0.gguf",
          "sha": null,
          "size": 5194117824
        },
        {
          "name": "gemma-4-E4B-it-SDFT_Heretic_RP.i1-Q4_1.gguf",
          "sha": null,
          "size": 5435945664
        },
        {
          "name": "gemma-4-E4B-it-SDFT_Heretic_RP.i1-Q4_K_M.gguf",
          "sha": null,
          "size": 5335286464
        },
        {
          "name": "gemma-4-E4B-it-SDFT_Heretic_RP.i1-Q4_K_S.gguf",
          "sha": null,
          "size": 5202965184
        },
        {
          "name": "gemma-4-E4B-it-SDFT_Heretic_RP.i1-Q5_K_M.gguf",
          "sha": null,
          "size": 5762908864
        },
        {
          "name": "gemma-4-E4B-it-SDFT_Heretic_RP.i1-Q5_K_S.gguf",
          "sha": null,
          "size": 5685965504
        },
        {
          "name": "gemma-4-E4B-it-SDFT_Heretic_RP.i1-Q6_K.gguf",
          "sha": null,
          "size": 6217257664
        },
        {
          "name": "gemma-4-E4B-it-SDFT_Heretic_RP.imatrix.gguf",
          "sha": null,
          "size": 4429024
        }
      ],
      "siblings_count": 18,
      "tags": [
        "transformers",
        "gguf",
        "gemma4",
        "roleplay",
        "rp",
        "merged",
        "sdft",
        "gguf-ready",
        "en",
        "base_model:Ilya626/gemma-4-E4B-it-SDFT_Heretic_RP",
        "base_model:quantized:Ilya626/gemma-4-E4B-it-SDFT_Heretic_RP",
        "license:apache-2.0",
        "endpoints_compatible",
        "region:us",
        "imatrix",
        "conversational"
      ],
      "trust_score": 50
    },
    "input_type": "huggingface",
    "repo_id": "mradermacher/gemma-4-E4B-it-SDFT_Heretic_RP-i1-GGUF",
    "repo_sha": "a6356be002df20b7c3abaa96b90b0ff328e2b0cf",
    "size_bytes": 4714691264
  },
  "tensors": {
    "count": 720,
    "dtype_histogram": {
      "F16": 1,
      "F32": 339,
      "IQ3_S": 289,
      "Q4_K": 89,
      "Q6_K": 2
    },
    "dtype_stats": {
      "F16": {
        "avg_bits_per_elem": 16.0,
        "bytes": 55050240,
        "count": 1,
        "mb": 52.5
      },
      "F32": {
        "avg_bits_per_elem": 32.016628,
        "bytes": 2264384,
        "count": 339,
        "mb": 2.16
      },
      "IQ3_S": {
        "avg_bits_per_elem": 3.4375,
        "bytes": 1524582400,
        "count": 289,
        "mb": 1453.96
      },
      "Q4_K": {
        "avg_bits_per_elem": 4.5,
        "bytes": 254361600,
        "count": 89,
        "mb": 242.58
      },
      "Q6_K": {
        "avg_bits_per_elem": 6.5625,
        "bytes": 2862612480,
        "count": 2,
        "mb": 2730.0
      }
    },
    "integrity": {
      "header_kv_dir_bytes": 15820160,
      "misaligned_tensors": 0,
      "offsets_monotone": true,
      "oob_tensor_count": 0,
      "overlapping_tensor_pairs": 0,
      "tensor_bytes_sum": 4698871104,
      "tensor_delta_bytes": 15820160,
      "tensor_delta_mb": 15.09,
      "trailing_bytes": 0,
      "unaccounted_bytes": 0
    },
    "preview": [
      {
        "bits_per_elem": 32.0,
        "byte_len": 10240,
        "dims": [
          2560
        ],
        "dtype_id": 0,
        "dtype_name": "F32",
        "index": 0,
        "name": "output_norm.weight",
        "offset": 0
      },
      {
        "bits_per_elem": 16.0,
        "byte_len": 55050240,
        "dims": [
          2560,
          10752
        ],
        "dtype_id": 1,
        "dtype_name": "F16",
        "index": 1,
        "name": "per_layer_model_proj.weight",
        "offset": 10240
      },
      {
        "bits_per_elem": 32.0,
        "byte_len": 1024,
        "dims": [
          256
        ],
        "dtype_id": 0,
        "dtype_name": "F32",
        "index": 2,
        "name": "per_layer_proj_norm.weight",
        "offset": 55060480
      },
      {
        "bits_per_elem": 6.5625,
        "byte_len": 2312110080,
        "dims": [
          10752,
          262144
        ],
        "dtype_id": 14,
        "dtype_name": "Q6_K",
        "index": 3,
        "name": "per_layer_token_embd.weight",
        "offset": 55061504
      },
      {
        "bits_per_elem": 32.0,
        "byte_len": 1024,
        "dims": [
          256
        ],
        "dtype_id": 0,
        "dtype_name": "F32",
        "index": 4,
        "name": "rope_freqs.weight",
        "offset": 2367171584
      },
      {
        "bits_per_elem": 6.5625,
        "byte_len": 550502400,
        "dims": [
          2560,
          262144
        ],
        "dtype_id": 14,
        "dtype_name": "Q6_K",
        "index": 5,
        "name": "token_embd.weight",
        "offset": 2367172608
      },
      {
        "bits_per_elem": 3.4375,
        "byte_len": 563200,
        "dims": [
          2560,
          512
        ],
        "dtype_id": 21,
        "dtype_name": "IQ3_S",
        "index": 6,
        "name": "blk.0.attn_k.weight",
        "offset": 2917675008
      },
      {
        "bits_per_elem": 32.0,
        "byte_len": 1024,
        "dims": [
          256
        ],
        "dtype_id": 0,
        "dtype_name": "F32",
        "index": 7,
        "name": "blk.0.attn_k_norm.weight",
        "offset": 2918238208
      },
      {
        "bits_per_elem": 32.0,
        "byte_len": 10240,
        "dims": [
          2560
        ],
        "dtype_id": 0,
        "dtype_name": "F32",
        "index": 8,
        "name": "blk.0.attn_norm.weight",
        "offset": 2918239232
      },
      {
        "bits_per_elem": 4.5,
        "byte_len": 2949120,
        "dims": [
          2048,
          2560
        ],
        "dtype_id": 12,
        "dtype_name": "Q4_K",
        "index": 9,
        "name": "blk.0.attn_output.weight",
        "offset": 2918249472
      },
      {
        "bits_per_elem": 3.4375,
        "byte_len": 2252800,
        "dims": [
          2560,
          2048
        ],
        "dtype_id": 21,
        "dtype_name": "IQ3_S",
        "index": 10,
        "name": "blk.0.attn_q.weight",
        "offset": 2921198592
      },
      {
        "bits_per_elem": 32.0,
        "byte_len": 1024,
        "dims": [
          256
        ],
        "dtype_id": 0,
        "dtype_name": "F32",
        "index": 11,
        "name": "blk.0.attn_q_norm.weight",
        "offset": 2923451392
      },
      {
        "bits_per_elem": 4.5,
        "byte_len": 737280,
        "dims": [
          2560,
          512
        ],
        "dtype_id": 12,
        "dtype_name": "Q4_K",
        "index": 12,
        "name": "blk.0.attn_v.weight",
        "offset": 2923452416
      },
      {
        "bits_per_elem": 4.5,
        "byte_len": 14745600,
        "dims": [
          10240,
          2560
        ],
        "dtype_id": 12,
        "dtype_name": "Q4_K",
        "index": 13,
        "name": "blk.0.ffn_down.weight",
        "offset": 2924189696
      },
      {
        "bits_per_elem": 3.4375,
        "byte_len": 11264000,
        "dims": [
          2560,
          10240
        ],
        "dtype_id": 21,
        "dtype_name": "IQ3_S",
        "index": 14,
        "name": "blk.0.ffn_gate.weight",
        "offset": 2938935296
      },
      {
        "bits_per_elem": 32.0,
        "byte_len": 10240,
        "dims": [
          2560
        ],
        "dtype_id": 0,
        "dtype_name": "F32",
        "index": 15,
        "name": "blk.0.ffn_norm.weight",
        "offset": 2950199296
      },
      {
        "bits_per_elem": 3.4375,
        "byte_len": 11264000,
        "dims": [
          2560,
          10240
        ],
        "dtype_id": 21,
        "dtype_name": "IQ3_S",
        "index": 16,
        "name": "blk.0.ffn_up.weight",
        "offset": 2950209536
      },
      {
        "bits_per_elem": 3.4375,
        "byte_len": 281600,
        "dims": [
          2560,
          256
        ],
        "dtype_id": 21,
        "dtype_name": "IQ3_S",
        "index": 17,
        "name": "blk.0.inp_gate.weight",
        "offset": 2961473536
      },
      {
        "bits_per_elem": 256.0,
        "byte_len": 32,
        "dims": [
          1
        ],
        "dtype_id": 0,
        "dtype_name": "F32",
        "index": 18,
        "name": "blk.0.layer_output_scale.weight",
        "offset": 2961755136
      },
      {
        "bits_per_elem": 32.0,
        "byte_len": 10240,
        "dims": [
          2560
        ],
        "dtype_id": 0,
        "dtype_name": "F32",
        "index": 19,
        "name": "blk.0.post_attention_norm.weight",
        "offset": 2961755168
      },
      {
        "bits_per_elem": 32.0,
        "byte_len": 10240,
        "dims": [
          2560
        ],
        "dtype_id": 0,
        "dtype_name": "F32",
        "index": 20,
        "name": "blk.0.post_ffw_norm.weight",
        "offset": 2961765408
      },
      {
        "bits_per_elem": 32.0,
        "byte_len": 10240,
        "dims": [
          2560
        ],
        "dtype_id": 0,
        "dtype_name": "F32",
        "index": 21,
        "name": "blk.0.post_norm.weight",
        "offset": 2961775648
      },
      {
        "bits_per_elem": 3.4375,
        "byte_len": 281600,
        "dims": [
          256,
          2560
        ],
        "dtype_id": 21,
        "dtype_name": "IQ3_S",
        "index": 22,
        "name": "blk.0.proj.weight",
        "offset": 2961785888
      },
      {
        "bits_per_elem": 3.4375,
        "byte_len": 563200,
        "dims": [
          2560,
          512
        ],
        "dtype_id": 21,
        "dtype_name": "IQ3_S",
        "index": 23,
        "name": "blk.1.attn_k.weight",
        "offset": 2962067488
      },
      {
        "bits_per_elem": 32.0,
        "byte_len": 1024,
        "dims": [
          256
        ],
        "dtype_id": 0,
        "dtype_name": "F32",
        "index": 24,
        "name": "blk.1.attn_k_norm.weight",
        "offset": 2962630688
      },
      {
        "bits_per_elem": 32.0,
        "byte_len": 10240,
        "dims": [
          2560
        ],
        "dtype_id": 0,
        "dtype_name": "F32",
        "index": 25,
        "name": "blk.1.attn_norm.weight",
        "offset": 2962631712
      },
      {
        "bits_per_elem": 4.5,
        "byte_len": 2949120,
        "dims": [
          2048,
          2560
        ],
        "dtype_id": 12,
        "dtype_name": "Q4_K",
        "index": 26,
        "name": "blk.1.attn_output.weight",
        "offset": 2962641952
      },
      {
        "bits_per_elem": 3.4375,
        "byte_len": 2252800,
        "dims": [
          2560,
          2048
        ],
        "dtype_id": 21,
        "dtype_name": "IQ3_S",
        "index": 27,
        "name": "blk.1.attn_q.weight",
        "offset": 2965591072
      },
      {
        "bits_per_elem": 32.0,
        "byte_len": 1024,
        "dims": [
          256
        ],
        "dtype_id": 0,
        "dtype_name": "F32",
        "index": 28,
        "name": "blk.1.attn_q_norm.weight",
        "offset": 2967843872
      },
      {
        "bits_per_elem": 4.5,
        "byte_len": 737280,
        "dims": [
          2560,
          512
        ],
        "dtype_id": 12,
        "dtype_name": "Q4_K",
        "index": 29,
        "name": "blk.1.attn_v.weight",
        "offset": 2967844896
      },
      {
        "bits_per_elem": 4.5,
        "byte_len": 14745600,
        "dims": [
          10240,
          2560
        ],
        "dtype_id": 12,
        "dtype_name": "Q4_K",
        "index": 30,
        "name": "blk.1.ffn_down.weight",
        "offset": 2968582176
      },
      {
        "bits_per_elem": 3.4375,
        "byte_len": 11264000,
        "dims": [
          2560,
          10240
        ],
        "dtype_id": 21,
        "dtype_name": "IQ3_S",
        "index": 31,
        "name": "blk.1.ffn_gate.weight",
        "offset": 2983327776
      },
      {
        "bits_per_elem": 32.0,
        "byte_len": 10240,
        "dims": [
          2560
        ],
        "dtype_id": 0,
        "dtype_name": "F32",
        "index": 32,
        "name": "blk.1.ffn_norm.weight",
        "offset": 2994591776
      },
      {
        "bits_per_elem": 3.4375,
        "byte_len": 11264000,
        "dims": [
          2560,
          10240
        ],
        "dtype_id": 21,
        "dtype_name": "IQ3_S",
        "index": 33,
        "name": "blk.1.ffn_up.weight",
        "offset": 2994602016
      },
      {
        "bits_per_elem": 3.4375,
        "byte_len": 281600,
        "dims": [
          2560,
          256
        ],
        "dtype_id": 21,
        "dtype_name": "IQ3_S",
        "index": 34,
        "name": "blk.1.inp_gate.weight",
        "offset": 3005866016
      },
      {
        "bits_per_elem": 256.0,
        "byte_len": 32,
        "dims": [
          1
        ],
        "dtype_id": 0,
        "dtype_name": "F32",
        "index": 35,
        "name": "blk.1.layer_output_scale.weight",
        "offset": 3006147616
      },
      {
        "bits_per_elem": 32.0,
        "byte_len": 10240,
        "dims": [
          2560
        ],
        "dtype_id": 0,
        "dtype_name": "F32",
        "index": 36,
        "name": "blk.1.post_attention_norm.weight",
        "offset": 3006147648
      },
      {
        "bits_per_elem": 32.0,
        "byte_len": 10240,
        "dims": [
          2560
        ],
        "dtype_id": 0,
        "dtype_name": "F32",
        "index": 37,
        "name": "blk.1.post_ffw_norm.weight",
        "offset": 3006157888
      },
      {
        "bits_per_elem": 32.0,
        "byte_len": 10240,
        "dims": [
          2560
        ],
        "dtype_id": 0,
        "dtype_name": "F32",
        "index": 38,
        "name": "blk.1.post_norm.weight",
        "offset": 3006168128
      },
      {
        "bits_per_elem": 3.4375,
        "byte_len": 281600,
        "dims": [
          256,
          2560
        ],
        "dtype_id": 21,
        "dtype_name": "IQ3_S",
        "index": 39,
        "name": "blk.1.proj.weight",
        "offset": 3006178368
      },
      {
        "bits_per_elem": 3.4375,
        "byte_len": 563200,
        "dims": [
          2560,
          512
        ],
        "dtype_id": 21,
        "dtype_name": "IQ3_S",
        "index": 40,
        "name": "blk.2.attn_k.weight",
        "offset": 3006459968
      },
      {
        "bits_per_elem": 32.0,
        "byte_len": 1024,
        "dims": [
          256
        ],
        "dtype_id": 0,
        "dtype_name": "F32",
        "index": 41,
        "name": "blk.2.attn_k_norm.weight",
        "offset": 3007023168
      },
      {
        "bits_per_elem": 32.0,
        "byte_len": 10240,
        "dims": [
          2560
        ],
        "dtype_id": 0,
        "dtype_name": "F32",
        "index": 42,
        "name": "blk.2.attn_norm.weight",
        "offset": 3007024192
      },
      {
        "bits_per_elem": 4.5,
        "byte_len": 2949120,
        "dims": [
          2048,
          2560
        ],
        "dtype_id": 12,
        "dtype_name": "Q4_K",
        "index": 43,
        "name": "blk.2.attn_output.weight",
        "offset": 3007034432
      },
      {
        "bits_per_elem": 3.4375,
        "byte_len": 2252800,
        "dims": [
          2560,
          2048
        ],
        "dtype_id": 21,
        "dtype_name": "IQ3_S",
        "index": 44,
        "name": "blk.2.attn_q.weight",
        "offset": 3009983552
      },
      {
        "bits_per_elem": 32.0,
        "byte_len": 1024,
        "dims": [
          256
        ],
        "dtype_id": 0,
        "dtype_name": "F32",
        "index": 45,
        "name": "blk.2.attn_q_norm.weight",
        "offset": 3012236352
      },
      {
        "bits_per_elem": 4.5,
        "byte_len": 737280,
        "dims": [
          2560,
          512
        ],
        "dtype_id": 12,
        "dtype_name": "Q4_K",
        "index": 46,
        "name": "blk.2.attn_v.weight",
        "offset": 3012237376
      },
      {
        "bits_per_elem": 4.5,
        "byte_len": 14745600,
        "dims": [
          10240,
          2560
        ],
        "dtype_id": 12,
        "dtype_name": "Q4_K",
        "index": 47,
        "name": "blk.2.ffn_down.weight",
        "offset": 3012974656
      },
      {
        "bits_per_elem": 3.4375,
        "byte_len": 11264000,
        "dims": [
          2560,
          10240
        ],
        "dtype_id": 21,
        "dtype_name": "IQ3_S",
        "index": 48,
        "name": "blk.2.ffn_gate.weight",
        "offset": 3027720256
      },
      {
        "bits_per_elem": 32.0,
        "byte_len": 10240,
        "dims": [
          2560
        ],
        "dtype_id": 0,
        "dtype_name": "F32",
        "index": 49,
        "name": "blk.2.ffn_norm.weight",
        "offset": 3038984256
      }
    ]
  },
  "tokenizer": {
    "bos_token_id": 2,
    "chat_template_chars": 11926,
    "eos_token_id": 1,
    "has_chat_template": true,
    "type": "gemma4",
    "vocab_size": 262144
  }
}