| { |
| "mic": "0.2", |
| "name": "SmolLM text generation", |
| "usage": "AutoTokenizer + AutoModelForCausalLM.generate, batched, left padding", |
| "architecture": "LlamaForCausalLM", |
| "checkpoints": { |
| "HuggingFaceTB/SmolLM-135M": { |
| "fixture": "hf-internal-testing/mic-SmolLM-slim" |
| }, |
| "HuggingFaceTB/SmolLM-360M": { |
| "fixture": "hf-internal-testing/mic-SmolLM-slim" |
| }, |
| "HuggingFaceTB/SmolLM-1.7B": { |
| "fixture": "hf-internal-testing/mic-SmolLM-1.7B-slim" |
| } |
| }, |
| "fixtures": { |
| "hf-internal-testing/mic-SmolLM-slim": { |
| "built_from": "HuggingFaceTB/SmolLM-135M", |
| "covers": [ |
| "HuggingFaceTB/SmolLM-135M", |
| "HuggingFaceTB/SmolLM-360M" |
| ], |
| "build": { |
| "seed": 17, |
| "overfit": "random weights generate the same tokens for every prompt, so generated_tokens cannot tell the prompts apart. The training rows share 'The capital of' and then name four countries, so ln 4 at that position (about 0.2 per token) cannot be learned: target_loss 0.3", |
| "target_loss": 0.3 |
| } |
| }, |
| "hf-internal-testing/mic-SmolLM-1.7B-slim": { |
| "built_from": "HuggingFaceTB/SmolLM-1.7B", |
| "covers": [ |
| "HuggingFaceTB/SmolLM-1.7B" |
| ], |
| "why_separate": "different tokenizer.json and tokenizer_config.json; full multi-head attention instead of grouped key/value heads" |
| } |
| }, |
| "integrations": { |
| "transformers": { |
| "entry_point": "integrations/transformers/usage.md", |
| "format": "runnable-markdown", |
| "requires": [ |
| "transformers>=5", |
| "torch", |
| "accelerate" |
| ], |
| "targets": [ |
| "fixtures", |
| "checkpoints" |
| ], |
| "variants": { |
| "default": {}, |
| "eager": { |
| "attn_implementation": "eager" |
| }, |
| "bfloat16": { |
| "dtype": "bfloat16" |
| } |
| }, |
| "training": true, |
| "resources": { |
| "fixture": { |
| "memory_gb": 1, |
| "download_mb": 16 |
| }, |
| "HuggingFaceTB/SmolLM-135M": { |
| "memory_gb": 2, |
| "download_mb": 513 |
| }, |
| "HuggingFaceTB/SmolLM-360M": { |
| "memory_gb": 3, |
| "download_mb": 1300 |
| }, |
| "HuggingFaceTB/SmolLM-1.7B": { |
| "memory_gb": 12, |
| "disk_gb": 8, |
| "download_mb": 6400, |
| "measured": "peak RSS 10.6 GB (default) and 11.3 GB (bfloat16: the variant reload holds two copies), one process, Apple M5, 2026-10-07" |
| } |
| } |
| }, |
| "transformers.js": { |
| "entry_point": "integrations/transformers.js/integration.js", |
| "format": "script", |
| "requires": [ |
| "@huggingface/transformers@4.3.0" |
| ], |
| "targets": [ |
| "checkpoints" |
| ], |
| "artifacts": { |
| "HuggingFaceTB/SmolLM-135M": { |
| "repo": "onnx-community/SmolLM-135M-ONNX", |
| "file": "onnx/model.onnx" |
| }, |
| "HuggingFaceTB/SmolLM-360M": { |
| "repo": "onnx-community/SmolLM-360M-ONNX", |
| "file": "onnx/model.onnx" |
| } |
| }, |
| "unsupported": { |
| "fixtures": "no ONNX conversion of the fixtures yet", |
| "HuggingFaceTB/SmolLM-1.7B": "no onnx-community conversion; the creator's float32 ONNX is 6.8 GB" |
| }, |
| "variants": { |
| "default": { |
| "dtype": "fp32" |
| }, |
| "webgpu": { |
| "dtype": "fp32", |
| "device": "webgpu" |
| }, |
| "webgpu-fp16": { |
| "dtype": "fp16", |
| "device": "webgpu" |
| }, |
| "webgpu-q4f16": { |
| "dtype": "q4f16", |
| "device": "webgpu" |
| } |
| }, |
| "training": false |
| } |
| }, |
| "outputs": { |
| "runtime": { |
| "compare": "exact" |
| }, |
| "generated_tokens": { |
| "compare": "exact" |
| }, |
| "generated_text": { |
| "compare": "exact" |
| }, |
| "next_token_logits": { |
| "compare": "allclose", |
| "atol": 0.0001, |
| "rtol": 0.0001 |
| }, |
| "training.loss": { |
| "compare": "allclose", |
| "atol": 0.0001, |
| "rtol": 0.0001 |
| }, |
| "training.gradients": { |
| "compare": "all_true" |
| }, |
| "training.updated": { |
| "compare": "true" |
| } |
| } |
| } |
|
|
