| { |
| "model": "qwen3_0_6b_q4gsw", |
| "upstream": { |
| "model": "Qwen/Qwen3-0.6B", |
| "revision": "c1899de289a04d12100db370d81485cdf75e47ca", |
| "official": true, |
| "safetensors_sha256": { |
| "model.safetensors": "f47f71177f32bcd101b7573ec9171e6a57f4f4d31148d38e382306f42996874b" |
| } |
| }, |
| "versions": { |
| "python": "3.12.15+meta", |
| "torch": "2.14.1+cpu", |
| "torchao": "0.18.0+cpu", |
| "executorch": "1.5.1+cpu", |
| "executorch_git": "3b60683923245cf472b7323426920e15623ba361", |
| "transformers": "5.19.0", |
| "huggingface_hub": "2.2.0" |
| }, |
| "export": { |
| "api": "python -m executorch.extension.llm.export.export_llm (hydra overrides below)", |
| "overrides": [ |
| "base.model_class=qwen3_0_6b", |
| "base.checkpoint=<out>/_work/checkpoint.pth", |
| "base.params=<out>/_work/params.json", |
| "model.use_kv_cache=true", |
| "model.use_sdpa_with_kv_cache=true", |
| "model.enable_dynamic_shape=true", |
| "model.dtype_override=fp32", |
| "export.max_seq_length=513", |
| "export.max_context_length=8960", |
| "export.output_dir=<out>", |
| "export.output_name=<out>/qwen3_0_6b_q4gsw.pte", |
| "quantization.qmode=4w", |
| "quantization.group_size=64", |
| "quantization.embedding_quantize='4,64'", |
| "backend.vulkan.enabled=true" |
| ], |
| "checkpoint_conversion": "executorch.examples.models.qwen3.convert_weights.convert_weights(<snapshot>, checkpoint.pth)", |
| "import_shim": "recipes/_shims/torchtune (re-exports executorch.examples.models.checkpoint.get_mapped_key; torchtune itself is not installed)", |
| "params": { |
| "dim": 1024, |
| "ffn_dim_multiplier": 1, |
| "hidden_dim": 3072, |
| "n_heads": 16, |
| "head_dim": 128, |
| "n_kv_heads": 8, |
| "n_layers": 28, |
| "norm_eps": 1e-06, |
| "rope_theta": 1000000.0, |
| "use_scaled_rope": false, |
| "vocab_size": 151936, |
| "use_hf_rope": true, |
| "attention_qkv_bias": false, |
| "use_qk_norm": true, |
| "qk_norm_before_rope": true |
| }, |
| "partitioner": "VulkanPartitioner via backend.vulkan.enabled (WebGPUPartitioner is a pass-through wrapper of it; delegate id VulkanBackend)", |
| "inputs": [ |
| { |
| "name": "tokens", |
| "shape": [ |
| 1, |
| "1..512" |
| ], |
| "dtype": "int64" |
| }, |
| { |
| "name": "input_pos", |
| "shape": [ |
| 1 |
| ], |
| "dtype": "int64" |
| } |
| ], |
| "outputs": [ |
| { |
| "name": "logits", |
| "shape": [ |
| 1, |
| 151936 |
| ], |
| "dtype": "float32", |
| "note": "last position only" |
| } |
| ], |
| "prefill_max": 512, |
| "max_seq_len_metadata": 513, |
| "max_context_length": 8960, |
| "dtype": "float32 activations", |
| "quantization": "4-bit groupwise weight-only linears (torchao IntxWeightOnlyConfig int4, group 64) and 4-bit groupwise embedding (group 64); output head quantized too", |
| "kv_cache": "fp32 buffers in the program; fp16 KV is a runtime option of the WebGPU runtime (et_set_kv_f16), not an export flag" |
| }, |
| "files": { |
| "qwen3_0_6b_q4gsw.pte": { |
| "bytes": 400532224, |
| "sha256": "057af5734a6ce17cbf19598bcc0e972aed2401d1522eed41249675c1ab372ce6" |
| }, |
| "tokenizer/tokenizer.json": { |
| "sha256": "aeb13307a71acd8fe81861d94ad54ab689df773318809eed3cbe794b4492dae4" |
| }, |
| "tokenizer/tokenizer_config.json": { |
| "sha256": "d5d09f07b48c3086c508b30d1c9114bd1189145b74e982a265350c923acd8101" |
| }, |
| "tokenizer/vocab.json": { |
| "sha256": "ca10d7e9fb3ed18575dd1e277a2579c16d108e32f27439684afa0e10b1440910" |
| }, |
| "tokenizer/merges.txt": { |
| "sha256": "8831e4f1a044471340f7c0a83d7bd71306a5b867e95fd870f74d0c5308a904d5" |
| }, |
| "tokenizer/generation_config.json": { |
| "sha256": "2325da0f15bb848e018c5ae071b7943332e9f871d6b60e2ed22ca97d4cb993d2" |
| }, |
| "tokenizer/config.json": { |
| "sha256": "660db3b73d788119c04535e48cf9be5f55bc3100841a718637ae695b442f27dd" |
| }, |
| "tokenizer/LICENSE": { |
| "sha256": "832dd9e00a68dd83b3c3fb9f5588dad7dcf337a0db50f7d9483f310cd292e92e" |
| } |
| }, |
| "graph": { |
| "non_delegated_kernels": {}, |
| "delegates": [ |
| { |
| "id": "VulkanBackend", |
| "ops": { |
| "et_vk.prepack.default": 2, |
| "sym_size.int": 1, |
| "et_vk.embedding_q4gsw.default": 1, |
| "et_vk.select_as_symint.default": 2, |
| "et_vk.rms_norm.default": 113, |
| "et_vk.linear_q4gsw.default": 197, |
| "aten.view_copy.default": 224, |
| "et_vk.apply_rotary_emb_hf.default": 28, |
| "sdpa_with_kv_cache.default": 28, |
| "aten.add.Tensor": 56, |
| "aten.sigmoid.default": 28, |
| "aten.mul.Tensor": 56, |
| "aten.select_copy.int": 1 |
| }, |
| "inputs": 2, |
| "outputs": 1 |
| } |
| ] |
| }, |
| "command": "python recipes/qwen3-0.6b.py --out out/qwen3-0.6b" |
| } |
|
|
