Hugging Face's logo

executorch-community
/
Qwen2.5-0.5B-Instruct-ExecuTorch-WebGPU

Text Generation
ExecuTorch
webgpu
on-device
browser
Model card Files
xet
 Community
Qwen2.5-0.5B-Instruct-ExecuTorch-WebGPU
File size: 4,822 Bytes
2c717f6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
{
  "model": "qwen2_5_0_5b_instruct_q4gsw",
  "upstream": {
    "model": "Qwen/Qwen2.5-0.5B-Instruct",
    "revision": "7ae557604adf67be50417f59c2c2f167def9a775",
    "official": true,
    "safetensors_sha256": {
      "model.safetensors": "fdf756fa7fcbe7404d5c60e26bff1a0c8b8aa1f72ced49e7dd0210fe288fb7fe"
    }
  },
  "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=qwen2_5_0_5b",
      "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=8192",
      "export.output_dir=<out>",
      "export.output_name=<out>/qwen2_5_0_5b_instruct_q4gsw.pte",
      "quantization.qmode=4w",
      "quantization.group_size=64",
      "quantization.embedding_quantize='4,64'",
      "backend.vulkan.enabled=true"
    ],
    "checkpoint_conversion": "executorch.examples.models.qwen2_5.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": 896,
      "ffn_dim_multiplier": 1,
      "hidden_dim": 4864,
      "n_heads": 14,
      "n_kv_heads": 2,
      "n_layers": 24,
      "norm_eps": 1e-06,
      "rope_theta": 1000000.0,
      "use_scaled_rope": false,
      "vocab_size": 151936,
      "use_hf_rope": true,
      "attention_qkv_bias": 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": 8192,
    "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": {
    "qwen2_5_0_5b_instruct_q4gsw.pte": {
      "bytes": 299447808,
      "sha256": "474274b934978f6c11a842d575a07417d6687eb4d93f6b12d30d144dce729faa"
    },
    "tokenizer/tokenizer.json": {
      "sha256": "c0382117ea329cdf097041132f6d735924b697924d6f6fc3945713e96ce87539"
    },
    "tokenizer/tokenizer_config.json": {
      "sha256": "5b5d4f65d0acd3b2d56a35b56d374a36cbc1c8fa5cf3b3febbbfabf22f359583"
    },
    "tokenizer/vocab.json": {
      "sha256": "ca10d7e9fb3ed18575dd1e277a2579c16d108e32f27439684afa0e10b1440910"
    },
    "tokenizer/merges.txt": {
      "sha256": "599bab54075088774b1733fde865d5bd747cbcc7a547c5bc12610e874e26f5e3"
    },
    "tokenizer/generation_config.json": {
      "sha256": "e558847a8b4402616f1273797b015104dc266fe4b520056fca88823ba8f8ebe6"
    },
    "tokenizer/config.json": {
      "sha256": "18e18afcaccafade98daf13a54092927904649e1dd4eba8299ab717d5d94ff45"
    },
    "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": 49,
          "et_vk.linear_q4gsw.default": 169,
          "aten.view_copy.default": 192,
          "et_vk.apply_rotary_emb_hf.default": 24,
          "sdpa_with_kv_cache.default": 24,
          "aten.add.Tensor": 48,
          "aten.sigmoid.default": 24,
          "aten.mul.Tensor": 48,
          "aten.select_copy.int": 1
        },
        "inputs": 2,
        "outputs": 1
      }
    ]
  },
  "command": "python recipes/qwen2.5-0.5b-instruct.py --out out/qwen2.5-0.5b-instruct"
}