Publish gemma-4-e2b-it-q4k-ehf16-af32 manifest update
Browse files
models/gemma-4-e2b-it-q4k-ehf16-af32/manifest.json
CHANGED
|
@@ -155,7 +155,9 @@
|
|
| 155 |
"retainQ4KMaterialization": false
|
| 156 |
},
|
| 157 |
"largeWeights": {
|
| 158 |
-
"gpuResidentOverrides": [
|
|
|
|
|
|
|
| 159 |
},
|
| 160 |
"execution": {
|
| 161 |
"kernels": {
|
|
@@ -167,12 +169,12 @@
|
|
| 167 |
"rmsnorm": {
|
| 168 |
"kernel": "rmsnorm.wgsl",
|
| 169 |
"entry": "main",
|
| 170 |
-
"digest": "sha256:
|
| 171 |
},
|
| 172 |
"final_norm_stable": {
|
| 173 |
"kernel": "rmsnorm.wgsl",
|
| 174 |
"entry": "main",
|
| 175 |
-
"digest": "sha256:
|
| 176 |
"precision": {
|
| 177 |
"inputDtype": "f32",
|
| 178 |
"outputDtype": "f32"
|
|
@@ -181,12 +183,12 @@
|
|
| 181 |
"gemv": {
|
| 182 |
"kernel": "matmul_gemv_subgroup.wgsl",
|
| 183 |
"entry": "main_vec4",
|
| 184 |
-
"digest": "sha256:
|
| 185 |
},
|
| 186 |
"rope": {
|
| 187 |
"kernel": "rope.wgsl",
|
| 188 |
"entry": "main",
|
| 189 |
-
"digest": "sha256:
|
| 190 |
},
|
| 191 |
"attn_decode": {
|
| 192 |
"kernel": "attention_decode_online_f16kv.wgsl",
|
|
@@ -212,7 +214,7 @@
|
|
| 212 |
"tiled": {
|
| 213 |
"kernel": "matmul_f16w_f32a.wgsl",
|
| 214 |
"entry": "main",
|
| 215 |
-
"digest": "sha256:
|
| 216 |
},
|
| 217 |
"q4_prefill_shared": {
|
| 218 |
"kernel": "fused_matmul_q4_batched_multicol_shared.wgsl",
|
|
@@ -222,7 +224,7 @@
|
|
| 222 |
"attn_stream": {
|
| 223 |
"kernel": "attention_streaming_f16kv.wgsl",
|
| 224 |
"entry": "main",
|
| 225 |
-
"digest": "sha256:
|
| 226 |
"precision": {
|
| 227 |
"kvDtype": "f16"
|
| 228 |
}
|
|
@@ -254,7 +256,7 @@
|
|
| 254 |
"lm_head_gemv": {
|
| 255 |
"kernel": "matmul_gemv_subgroup.wgsl",
|
| 256 |
"entry": "main_multicol",
|
| 257 |
-
"digest": "sha256:
|
| 258 |
"constants": {
|
| 259 |
"MULTICOL_COLS_PER_WG": 64,
|
| 260 |
"MULTICOL_THREADS_PER_COL": 4
|
|
@@ -263,7 +265,7 @@
|
|
| 263 |
"lm_head_gemv_stable": {
|
| 264 |
"kernel": "matmul_gemv_subgroup.wgsl",
|
| 265 |
"entry": "main_multicol",
|
| 266 |
-
"digest": "sha256:
|
| 267 |
"constants": {
|
| 268 |
"MULTICOL_COLS_PER_WG": 64,
|
| 269 |
"MULTICOL_THREADS_PER_COL": 4
|
|
@@ -276,7 +278,7 @@
|
|
| 276 |
"lm_head_prefill_stable": {
|
| 277 |
"kernel": "matmul_f16w_f32a.wgsl",
|
| 278 |
"entry": "main",
|
| 279 |
-
"digest": "sha256:
|
| 280 |
"precision": {
|
| 281 |
"inputDtype": "f32",
|
| 282 |
"outputDtype": "f32"
|
|
@@ -285,7 +287,7 @@
|
|
| 285 |
"sample": {
|
| 286 |
"kernel": "sample.wgsl",
|
| 287 |
"entry": "sample_single_pass",
|
| 288 |
-
"digest": "sha256:
|
| 289 |
}
|
| 290 |
},
|
| 291 |
"preLayer": [
|
|
|
|
| 155 |
"retainQ4KMaterialization": false
|
| 156 |
},
|
| 157 |
"largeWeights": {
|
| 158 |
+
"gpuResidentOverrides": [
|
| 159 |
+
"model.language_model.embed_tokens.weight"
|
| 160 |
+
]
|
| 161 |
},
|
| 162 |
"execution": {
|
| 163 |
"kernels": {
|
|
|
|
| 169 |
"rmsnorm": {
|
| 170 |
"kernel": "rmsnorm.wgsl",
|
| 171 |
"entry": "main",
|
| 172 |
+
"digest": "sha256:c723c0300cc8f93396d79d1a274c0a90ccf1eb8ce82c0d778b8cd4659615fe45"
|
| 173 |
},
|
| 174 |
"final_norm_stable": {
|
| 175 |
"kernel": "rmsnorm.wgsl",
|
| 176 |
"entry": "main",
|
| 177 |
+
"digest": "sha256:c723c0300cc8f93396d79d1a274c0a90ccf1eb8ce82c0d778b8cd4659615fe45",
|
| 178 |
"precision": {
|
| 179 |
"inputDtype": "f32",
|
| 180 |
"outputDtype": "f32"
|
|
|
|
| 183 |
"gemv": {
|
| 184 |
"kernel": "matmul_gemv_subgroup.wgsl",
|
| 185 |
"entry": "main_vec4",
|
| 186 |
+
"digest": "sha256:2ea9aea659367210fbf61d922c0d4ee17cdac68ffb506b3bc5516e498071e011"
|
| 187 |
},
|
| 188 |
"rope": {
|
| 189 |
"kernel": "rope.wgsl",
|
| 190 |
"entry": "main",
|
| 191 |
+
"digest": "sha256:e13c54f261d7ef00474b71e4fe7b3298fb5cc49161f50c7ea1854aa6672ebe71"
|
| 192 |
},
|
| 193 |
"attn_decode": {
|
| 194 |
"kernel": "attention_decode_online_f16kv.wgsl",
|
|
|
|
| 214 |
"tiled": {
|
| 215 |
"kernel": "matmul_f16w_f32a.wgsl",
|
| 216 |
"entry": "main",
|
| 217 |
+
"digest": "sha256:027a8f1cd9713cbe0b0ada160bd175e0542bb90896ad85441b023522d9a1befc"
|
| 218 |
},
|
| 219 |
"q4_prefill_shared": {
|
| 220 |
"kernel": "fused_matmul_q4_batched_multicol_shared.wgsl",
|
|
|
|
| 224 |
"attn_stream": {
|
| 225 |
"kernel": "attention_streaming_f16kv.wgsl",
|
| 226 |
"entry": "main",
|
| 227 |
+
"digest": "sha256:76eda2dfc1b4323ed4fd02b91d1ce448de65ce9b31898dee8ae635b1d5b94d3a",
|
| 228 |
"precision": {
|
| 229 |
"kvDtype": "f16"
|
| 230 |
}
|
|
|
|
| 256 |
"lm_head_gemv": {
|
| 257 |
"kernel": "matmul_gemv_subgroup.wgsl",
|
| 258 |
"entry": "main_multicol",
|
| 259 |
+
"digest": "sha256:e41f94574d5ac54dd2710036da5d5acc643a483b79b2d74b86825efdaaa7438f",
|
| 260 |
"constants": {
|
| 261 |
"MULTICOL_COLS_PER_WG": 64,
|
| 262 |
"MULTICOL_THREADS_PER_COL": 4
|
|
|
|
| 265 |
"lm_head_gemv_stable": {
|
| 266 |
"kernel": "matmul_gemv_subgroup.wgsl",
|
| 267 |
"entry": "main_multicol",
|
| 268 |
+
"digest": "sha256:e41f94574d5ac54dd2710036da5d5acc643a483b79b2d74b86825efdaaa7438f",
|
| 269 |
"constants": {
|
| 270 |
"MULTICOL_COLS_PER_WG": 64,
|
| 271 |
"MULTICOL_THREADS_PER_COL": 4
|
|
|
|
| 278 |
"lm_head_prefill_stable": {
|
| 279 |
"kernel": "matmul_f16w_f32a.wgsl",
|
| 280 |
"entry": "main",
|
| 281 |
+
"digest": "sha256:027a8f1cd9713cbe0b0ada160bd175e0542bb90896ad85441b023522d9a1befc",
|
| 282 |
"precision": {
|
| 283 |
"inputDtype": "f32",
|
| 284 |
"outputDtype": "f32"
|
|
|
|
| 287 |
"sample": {
|
| 288 |
"kernel": "sample.wgsl",
|
| 289 |
"entry": "sample_single_pass",
|
| 290 |
+
"digest": "sha256:7ff3dd675b4737d920e0b3889e15c4fedde8410a353bbc6d4651db76ce69eb9b"
|
| 291 |
}
|
| 292 |
},
|
| 293 |
"preLayer": [
|