clocksmith commited on
Commit
3f54bc5
·
verified ·
1 Parent(s): 440777e

Publish qwen-3-5-2b-q4k-ehaf16 manifest update

Browse files
models/qwen-3-5-2b-q4k-ehaf16/manifest.json CHANGED
@@ -128,10 +128,10 @@
128
  "session": {
129
  "compute": {
130
  "defaults": {
131
- "activationDtype": "f32",
132
- "mathDtype": "f32",
133
- "accumDtype": "f32",
134
- "outputDtype": "f32"
135
  }
136
  },
137
  "kvcache": {
@@ -194,66 +194,80 @@
194
  "digest": "sha256:257973dd762541d8a2c4b16ad8ad1d071199199e81e477ab644dce915961a989"
195
  },
196
  "rmsnorm": {
197
- "kernel": "rmsnorm.wgsl",
198
  "entry": "main",
199
- "digest": "sha256:f516b3e4bde2015f2a207c3ca5b8c9820c7809fa8f8d0786f90c568e0f1ac077"
200
  },
201
  "q4_decode": {
202
- "kernel": "fused_matmul_q4.wgsl",
203
- "entry": "main_multicol",
204
- "digest": "sha256:c0262226752914e863e8b5dc34479b523f9c93e2cef69c9057dcbde2b82854ef"
 
 
 
 
205
  },
206
  "rope": {
207
- "kernel": "rope.wgsl",
208
  "entry": "main",
209
- "digest": "sha256:4c803ad5e0dd065d5572c7aecc1def277c43884dcc02f22a9676914c10111400"
210
  },
211
  "attn_decode": {
212
- "kernel": "attention_decode_online_f16kv.wgsl",
213
  "entry": "main",
214
- "digest": "sha256:4c5d8c92a0a111af716d6b46b9559446807c086027445c7fefa150202f43dae4"
 
 
 
215
  },
216
  "residual": {
217
- "kernel": "residual.wgsl",
218
  "entry": "main",
219
- "digest": "sha256:f1abd88c959c5d8dd27b9353d487e37b2a96850ed9d90c365212e260399cc2a7"
220
  },
221
  "silu": {
222
- "kernel": "silu.wgsl",
223
  "entry": "main",
224
- "digest": "sha256:7b52d30fb741beef2dbf728e0c4ecffe5b08d9661d63c306caecb4cb3ced85e5",
225
  "constants": {
226
  "HAS_GATE": true
227
  }
228
  },
229
  "q4_prefill": {
230
- "kernel": "fused_matmul_q4_batched.wgsl",
231
- "entry": "main_batched",
232
- "digest": "sha256:a513e4170792aaf2f1c22b73f34288721541acf5954756993a0001a412c91209"
 
 
 
 
233
  },
234
  "tiled": {
235
- "kernel": "matmul_f16w_f32a.wgsl",
236
  "entry": "main",
237
- "digest": "sha256:027a8f1cd9713cbe0b0ada160bd175e0542bb90896ad85441b023522d9a1befc"
238
  },
239
  "attn_stream": {
240
- "kernel": "attention_streaming_f16kv.wgsl",
241
  "entry": "main",
242
- "digest": "sha256:b337a2dcf7b40431a733d1726eee8bf23504136fd5f915bec057fb59f3ea7480"
 
 
 
243
  },
244
  "lm_head_gemv": {
245
- "kernel": "matmul_gemv_subgroup.wgsl",
246
  "entry": "main_multicol",
247
- "digest": "sha256:96c38c15e6fed0d7efdc5cd094db5843a8e8ddfe01eee3bc7322fa555dacf3d0",
248
  "constants": {
249
  "MULTICOL_COLS_PER_WG": 64,
250
  "MULTICOL_THREADS_PER_COL": 4
251
  }
252
  },
253
  "sample": {
254
- "kernel": "sample.wgsl",
255
  "entry": "sample_single_pass",
256
- "digest": "sha256:4412357e84113ee2f1bc0dc8bf89e314c2ab482c89c14ca016ea9949d16a9d0c"
257
  }
258
  },
259
  "preLayer": [
@@ -4741,7 +4755,7 @@
4741
  "convertedAt": "2026-04-05T01:30:03.550Z",
4742
  "hasTokenizer": true,
4743
  "manifestRefresh": {
4744
- "at": "2026-04-08T01:20:32.365Z",
4745
  "config": "qwen-3-5-2b-q4k-ehaf16.json"
4746
  }
4747
  },
 
128
  "session": {
129
  "compute": {
130
  "defaults": {
131
+ "activationDtype": "f16",
132
+ "mathDtype": "f16",
133
+ "accumDtype": "f16",
134
+ "outputDtype": "f16"
135
  }
136
  },
137
  "kvcache": {
 
194
  "digest": "sha256:257973dd762541d8a2c4b16ad8ad1d071199199e81e477ab644dce915961a989"
195
  },
196
  "rmsnorm": {
197
+ "kernel": "rmsnorm_f16.wgsl",
198
  "entry": "main",
199
+ "digest": "sha256:6ff712c215fa8ea07b653ad51666620854b4ed03b2371b1e97ba2322efdfa665"
200
  },
201
  "q4_decode": {
202
+ "kernel": "fused_matmul_q4_multicol_f16a.wgsl",
203
+ "entry": "main_multicol_f16a",
204
+ "digest": "sha256:eb13c120d8c9829adc8bbd45cc44774e9be9f8f14c5e37ea68d747ad41e2eb4b",
205
+ "precision": {
206
+ "inputDtype": "f16",
207
+ "outputDtype": "f16"
208
+ }
209
  },
210
  "rope": {
211
+ "kernel": "rope_f16.wgsl",
212
  "entry": "main",
213
+ "digest": "sha256:26ab977df73442f99f3a70c8eb45681bfa12b3f8c4303313207aeb9c668ac044"
214
  },
215
  "attn_decode": {
216
+ "kernel": "attention_decode_online_f16.wgsl",
217
  "entry": "main",
218
+ "digest": "sha256:23ac23b33e1ca30bc6a0ebb90d1bcde0236604bf57217c1763186046e0fe03de",
219
+ "precision": {
220
+ "kvDtype": "f16"
221
+ }
222
  },
223
  "residual": {
224
+ "kernel": "residual_f16.wgsl",
225
  "entry": "main",
226
+ "digest": "sha256:5e843d5336d093da3c1bc042c39a570562e2744fce8e394bd57c3705076122c7"
227
  },
228
  "silu": {
229
+ "kernel": "silu_f16.wgsl",
230
  "entry": "main",
231
+ "digest": "sha256:24ce5e7aba948b074b57953e8d9279196231d9d26732f30bc1199ce979d39941",
232
  "constants": {
233
  "HAS_GATE": true
234
  }
235
  },
236
  "q4_prefill": {
237
+ "kernel": "fused_matmul_q4_batched_f16a.wgsl",
238
+ "entry": "main_batched_f16a",
239
+ "digest": "sha256:6f63a9c3b7c8b2ae8570987fb3314c1d8e5ab0b629c1ae1695ed2c45137c8564",
240
+ "precision": {
241
+ "inputDtype": "f16",
242
+ "outputDtype": "f16"
243
+ }
244
  },
245
  "tiled": {
246
+ "kernel": "matmul_f16.wgsl",
247
  "entry": "main",
248
+ "digest": "sha256:7e9f23f326b5750fc7783a1959334c237aff4a23da6f87176608c5c4eb4fc278"
249
  },
250
  "attn_stream": {
251
+ "kernel": "attention_small_f16.wgsl",
252
  "entry": "main",
253
+ "digest": "sha256:83f15e3d360f6425c13fc9fa9d0aab04c1de3616447de576eb0c5a6c89cfc4d0",
254
+ "precision": {
255
+ "kvDtype": "f16"
256
+ }
257
  },
258
  "lm_head_gemv": {
259
+ "kernel": "matmul_gemv_subgroup_f16a.wgsl",
260
  "entry": "main_multicol",
261
+ "digest": "sha256:cc31df534551ba5dd7feebbddb2822d1c53f7d6a98fd0d772c294620602b51e5",
262
  "constants": {
263
  "MULTICOL_COLS_PER_WG": 64,
264
  "MULTICOL_THREADS_PER_COL": 4
265
  }
266
  },
267
  "sample": {
268
+ "kernel": "sample_f16.wgsl",
269
  "entry": "sample_single_pass",
270
+ "digest": "sha256:ac0bacc17b3efb877b8b69a21a529f2ef2b21dd3307e3a7ce12f8cc0f785e275"
271
  }
272
  },
273
  "preLayer": [
 
4755
  "convertedAt": "2026-04-05T01:30:03.550Z",
4756
  "hasTokenizer": true,
4757
  "manifestRefresh": {
4758
+ "at": "2026-04-09T11:59:34.978Z",
4759
  "config": "qwen-3-5-2b-q4k-ehaf16.json"
4760
  }
4761
  },