DEMOThis report is public and visible to anyone with the link for 30 days, then deleted. Save privately — create free account
overlapping-tensors.safetensors
Format: SAFETENSORS  ·  Size: 2.1 KB  ·  Scanned: 2026-08-10  ·  1 ms
SHA256: 8ed50ec8cdb4027f2f63f46afeeb81233e7919e2…

Security grade

D
Score: 49/100
1 CRITICAL 1 MEDIUM 1 LOW 3 INFO
Artifact Security
D / 49
Repository Trust
N/A
local upload
Deployment Confidence
Low
Artifact Security grades the file. Repository Trust is unavailable for local uploads — no external provenance to verify. Deployment Confidence reflects artifact grade only.
Score breakdown
Base score 100
CRITICAL Overlapping tensor data regions (1 pairs) -40
LOW Sparse __metadata__: missing author, license, source/model_name -3
MEDIUM Misaligned tensor data (1 tensor(s)) -8
Penalty subtotal 49/100
Final D / 49
Safetensors unknown arch | F32 | 4 tensors | ~0M params | 0 MB | 1 blocking issue / 2 warnings / 3 informational

Runtime compatibility

Runtime Status Quant Architecture Context
HuggingFace Transformers
Direct
? Native format -- safetensors is the default weight format for HF models.
vLLM
Verify arch support
? ? ?
verify
vLLM targets decoder-only LLMs; this architecture may require a custom backend.
llama.cpp
Via conversion
? Requires GGUF conversion. Use convert_hf_to_gguf.py from the llama.cpp repository.
ONNX Runtime
Verify arch support
Via conversion
? ?
verify
Requires ONNX export (e.g., optimum-cli export onnx or torch.onnx.export).

Model info

Parameters ~
Estimated total number of weight values; determines VRAM needed at inference
448
Tensors
Total number of weight tensors stored in the file
4

Security findings (6)

CRITICAL Integrity
Overlapping tensor data regions (1 pairs)
Pairs: [('blk.0.attn_q.weight', 'blk.0.attn_k.weight')]
→ Overlapping regions enable aliasing attacks where two tensors share memory; reject this file
MEDIUM Memory Alignment
Misaligned tensor data (1 tensor(s))
Tensors not aligned to 8-byte boundary: ['blk.0.attn_k.weight']. Misalignment can cause SIGBUS on ARM and degrade SIMD performance on x86. Likely indicates manual file editing or a non-standard exporter.
→ Re-export using the official safetensors library to ensure correct alignment.
LOW Provenance
Sparse __metadata__: missing author, license, source/model_name
Embedded metadata has 1 key(s) but lacks: author, license, source/model_name. Without provenance, the model origin cannot be verified from the file alone.
→ Add author, license, and source fields to __metadata__ when exporting.
INFO Precision
Full-precision weights (F32) — not quantized
All 4 tensors use 32-bit float. Higher numerical fidelity but 2× the memory of F16 and 4-8× of quantized variants.
→ Consider converting to BF16 or INT8 for production inference if accuracy allows.
INFO Structure
Dtype breakdown: {'F32': 4}
4 tensors, ~448 parameters
→ Verify dtype matches expected model precision
INFO Format
Safetensors format provides strong sandboxing
Unlike pickle-based formats, safetensors cannot execute arbitrary code during loading
→ Safetensors is the recommended format for sharing model weights safely

Integrity & Structure

SHA-2568ed50ec8cdb4027f2f63f46afeeb81233e7919e2463a336452e42515092e20ff
SHA-512b6e2799c983f6f6a553ca30ac76d7901c84108ff9b9c2c3369abaee69af281244c982402c782e4c9…
File size2,182 bytes (0.0 MB)
Header size382 bytes (0.37 KB)
Tensor data sum1,792 bytes
Padding / overhead0.0 MB
Offsets monotone✓ OK
Overlapping regions⚠ 1

DType Mix

DTypeCountTotal bytesTotal MBAvg bits/elem
F32 4 1,792 0.0 32.0

Tensor Preview (first 4)

#NameShapeDTypeOffsetBytesbits/elem
0 token_embd.weight [32, 8] F32 0 1,024 32.0
1 blk.0.attn_q.weight [8, 8] F32 1,024 256 32.0
2 blk.0.attn_k.weight [8, 8] F32 1,028 256 32.0
3 blk.0.attn_v.weight [8, 8] F32 1,536 256 32.0

Model metadata (__metadata__)

formatpt

Scan coverage

Checks performed
  • Header JSON parse
  • Tensor offset monotonicity
  • Tensor overlap detection
  • Tensor out-of-bounds detection
  • Dtype histogram
  • Header size vs file size validation
  • Metadata key scan
  • Runtime compatibility heuristics
  • External model card / repository context: not available for local upload
  • Upstream repository: not checked — not available for direct file upload
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": true,
      "context": "Native format -- safetensors is the default weight format for HF models.",
      "direct_supported": true,
      "runtime": "HuggingFace Transformers"
    },
    {
      "arch": null,
      "context": "vLLM targets decoder-only LLMs; this architecture may require a custom backend.",
      "conversion_possible": null,
      "direct_supported": null,
      "runtime": "vLLM",
      "verification_required": true
    },
    {
      "arch": false,
      "context": "Requires GGUF conversion. Use convert_hf_to_gguf.py from the llama.cpp repository.",
      "conversion_possible": true,
      "direct_supported": false,
      "runtime": "llama.cpp"
    },
    {
      "arch": null,
      "context": "Requires ONNX export (e.g., optimum-cli export onnx or torch.onnx.export).",
      "conversion_possible": true,
      "direct_supported": false,
      "runtime": "ONNX Runtime"
    }
  ],
  "format": "LLM-SCAN",
  "format_version": "1.0",
  "generator": {
    "name": "llmscan-engine",
    "version": "2.0.0"
  },
  "model": {
    "arch": null,
    "dtype_histogram": {
      "F32": 4
    },
    "dtype_stats": {
      "F32": {
        "avg_bits_per_elem": 32.0,
        "bytes": 1792,
        "count": 4,
        "mb": 0.0
      }
    },
    "file_size_bytes": 2182,
    "file_size_mb": 0.0,
    "format": "safetensors",
    "has_metadata": true,
    "header_size_bytes": 382,
    "kv_meta": {},
    "metadata": {
      "format": "pt"
    },
    "offsets_monotone": true,
    "oob_tensor_count": 0,
    "overlapping_regions": 1,
    "param_estimate": 448,
    "params_estimate": 448,
    "quantization": null,
    "scanner_model_info": {
      "dtype_histogram": {
        "F32": 4
      },
      "dtype_percentage": {
        "F32": 100.0
      },
      "dtype_stats": {
        "F32": {
          "avg_bits_per_elem": 32.0,
          "bytes": 1792,
          "count": 4,
          "mb": 0.0
        }
      },
      "file_size_bytes": 2182,
      "file_size_mb": 0.0,
      "has_metadata": true,
      "header_size_bytes": 382,
      "inferred_arch": null,
      "metadata": {
        "format": "pt"
      },
      "nan_inf_scan": {
        "affected_tensors": [],
        "inf_count": 0,
        "method": "full",
        "nan_count": 0,
        "performed": true,
        "tensors_scanned": 4,
        "total_float_tensors": 4,
        "values_scanned": 448
      },
      "offsets_monotone": true,
      "oob_tensor_count": 0,
      "overlapping_regions": 1,
      "param_estimate": 448,
      "scalar_tensor_count": 0,
      "sha256": "8ed50ec8cdb4027f2f63f46afeeb81233e7919e2463a336452e42515092e20ff",
      "sha512": "b6e2799c983f6f6a553ca30ac76d7901c84108ff9b9c2c3369abaee69af281244c982402c782e4c96c789ae0973789f5b61b8a5bb0291c3d3847a28df3af0efc",
      "tensor_bytes_sum": 1792,
      "tensor_count": 4,
      "tensor_delta_bytes": 0,
      "tensor_delta_mb": 0.0,
      "tensor_preview": [
        {
          "bits_per_elem": 32.0,
          "byte_len": 1024,
          "dims": [
            32,
            8
          ],
          "dtype_name": "F32",
          "index": 0,
          "name": "token_embd.weight",
          "offset": 0
        },
        {
          "bits_per_elem": 32.0,
          "byte_len": 256,
          "dims": [
            8,
            8
          ],
          "dtype_name": "F32",
          "index": 1,
          "name": "blk.0.attn_q.weight",
          "offset": 1024
        },
        {
          "bits_per_elem": 32.0,
          "byte_len": 256,
          "dims": [
            8,
            8
          ],
          "dtype_name": "F32",
          "index": 2,
          "name": "blk.0.attn_k.weight",
          "offset": 1028
        },
        {
          "bits_per_elem": 32.0,
          "byte_len": 256,
          "dims": [
            8,
            8
          ],
          "dtype_name": "F32",
          "index": 3,
          "name": "blk.0.attn_v.weight",
          "offset": 1536
        }
      ],
      "total_bytes_declared": 1792,
      "unknown_dtypes": []
    },
    "sha256": "8ed50ec8cdb4027f2f63f46afeeb81233e7919e2463a336452e42515092e20ff",
    "sha512": "b6e2799c983f6f6a553ca30ac76d7901c84108ff9b9c2c3369abaee69af281244c982402c782e4c96c789ae0973789f5b61b8a5bb0291c3d3847a28df3af0efc",
    "tensor_bytes_sum": 1792,
    "tensor_count": 4,
    "tensor_delta_mb": 0.0,
    "tensor_preview": [
      {
        "bits_per_elem": 32.0,
        "byte_len": 1024,
        "dims": [
          32,
          8
        ],
        "dtype_name": "F32",
        "index": 0,
        "name": "token_embd.weight",
        "offset": 0
      },
      {
        "bits_per_elem": 32.0,
        "byte_len": 256,
        "dims": [
          8,
          8
        ],
        "dtype_name": "F32",
        "index": 1,
        "name": "blk.0.attn_q.weight",
        "offset": 1024
      },
      {
        "bits_per_elem": 32.0,
        "byte_len": 256,
        "dims": [
          8,
          8
        ],
        "dtype_name": "F32",
        "index": 2,
        "name": "blk.0.attn_k.weight",
        "offset": 1028
      },
      {
        "bits_per_elem": 32.0,
        "byte_len": 256,
        "dims": [
          8,
          8
        ],
        "dtype_name": "F32",
        "index": 3,
        "name": "blk.0.attn_v.weight",
        "offset": 1536
      }
    ]
  },
  "report": {
    "bullets": [
      "Safetensors unknown arch | F32 | 4 tensors | ~0M params | 0 MB | 1 blocking issue / 2 warnings / 3 informational",
      "SHA256: 8ed50ec8cdb4027f..."
    ],
    "recommendations": [
      "Overlapping regions enable aliasing attacks where two tensors share memory; reject this file",
      "Re-export using the official safetensors library to ensure correct alignment.",
      "Add author, license, and source fields to __metadata__ when exporting."
    ],
    "summary": "Safetensors unknown arch | F32 | 4 tensors | ~0M params | 0 MB | 1 blocking issue / 2 warnings / 3 informational"
  },
  "scan_context": {
    "duration_ms": 1,
    "ended_at": "2026-08-10T20:38:48+00:00Z",
    "limits": {
      "sandbox": true,
      "timeout_ms": 60000
    },
    "mode": "static-deep",
    "started_at": "2026-08-10T20:38:48+00:00Z"
  },
  "security": {
    "findings": [
      {
        "category": "Integrity",
        "detail": "Pairs: [(\u0027blk.0.attn_q.weight\u0027, \u0027blk.0.attn_k.weight\u0027)]",
        "recommendation": "Overlapping regions enable aliasing attacks where two tensors share memory; reject this file",
        "severity": "CRITICAL",
        "title": "Overlapping tensor data regions (1 pairs)"
      },
      {
        "category": "Precision",
        "detail": "All 4 tensors use 32-bit float. Higher numerical fidelity but 2\u00d7 the memory of F16 and 4-8\u00d7 of quantized variants.",
        "recommendation": "Consider converting to BF16 or INT8 for production inference if accuracy allows.",
        "severity": "INFO",
        "title": "Full-precision weights (F32) \u2014 not quantized"
      },
      {
        "category": "Provenance",
        "detail": "Embedded metadata has 1 key(s) but lacks: author, license, source/model_name. Without provenance, the model origin cannot be verified from the file alone.",
        "recommendation": "Add author, license, and source fields to __metadata__ when exporting.",
        "severity": "LOW",
        "title": "Sparse __metadata__: missing author, license, source/model_name"
      },
      {
        "category": "Structure",
        "detail": "4 tensors, ~448 parameters",
        "recommendation": "Verify dtype matches expected model precision",
        "severity": "INFO",
        "title": "Dtype breakdown: {\u0027F32\u0027: 4}"
      },
      {
        "category": "Format",
        "detail": "Unlike pickle-based formats, safetensors cannot execute arbitrary code during loading",
        "recommendation": "Safetensors is the recommended format for sharing model weights safely",
        "severity": "INFO",
        "title": "Safetensors format provides strong sandboxing"
      },
      {
        "category": "Memory Alignment",
        "detail": "Tensors not aligned to 8-byte boundary: [\u0027blk.0.attn_k.weight\u0027]. Misalignment can cause SIGBUS on ARM and degrade SIMD performance on x86. Likely indicates manual file editing or a non-standard exporter.",
        "recommendation": "Re-export using the official safetensors library to ensure correct alignment.",
        "severity": "MEDIUM",
        "title": "Misaligned tensor data (1 tensor(s))"
      }
    ],
    "grade": "D",
    "score": 49
  },
  "source": {
    "filename": "overlapping-tensors.safetensors",
    "hashes": {
      "sha256": "8ed50ec8cdb4027f2f63f46afeeb81233e7919e2463a336452e42515092e20ff"
    },
    "input_type": "file",
    "size_bytes": 2182
  },
  "tensors": {
    "count": 4,
    "dtype_histogram": {
      "F32": 4
    },
    "dtype_stats": {
      "F32": {
        "avg_bits_per_elem": 32.0,
        "bytes": 1792,
        "count": 4,
        "mb": 0.0
      }
    },
    "integrity": {
      "header_size_bytes": 382,
      "offsets_monotone": true,
      "oob_tensor_count": 0,
      "overlapping_regions": 1,
      "tensor_bytes_sum": 1792,
      "tensor_delta_mb": 0.0
    },
    "preview": [
      {
        "bits_per_elem": 32.0,
        "byte_len": 1024,
        "dims": [
          32,
          8
        ],
        "dtype_name": "F32",
        "index": 0,
        "name": "token_embd.weight",
        "offset": 0
      },
      {
        "bits_per_elem": 32.0,
        "byte_len": 256,
        "dims": [
          8,
          8
        ],
        "dtype_name": "F32",
        "index": 1,
        "name": "blk.0.attn_q.weight",
        "offset": 1024
      },
      {
        "bits_per_elem": 32.0,
        "byte_len": 256,
        "dims": [
          8,
          8
        ],
        "dtype_name": "F32",
        "index": 2,
        "name": "blk.0.attn_k.weight",
        "offset": 1028
      },
      {
        "bits_per_elem": 32.0,
        "byte_len": 256,
        "dims": [
          8,
          8
        ],
        "dtype_name": "F32",
        "index": 3,
        "name": "blk.0.attn_v.weight",
        "offset": 1536
      }
    ]
  },
  "tokenizer": {}
}