Publish gemma-3-1b-it-q4k-ehf16-af32 RDRR artifact
Browse files
models/gemma-3-1b-it-q4k-ehf16-af32/manifest.json
CHANGED
|
@@ -35,7 +35,7 @@
|
|
| 35 |
"valueNorm": false
|
| 36 |
},
|
| 37 |
"normalization": {
|
| 38 |
-
"rmsNormEps":
|
| 39 |
"rmsNormWeightOffset": true,
|
| 40 |
"postAttentionNorm": true,
|
| 41 |
"preFeedforwardNorm": true,
|
|
@@ -113,6 +113,29 @@
|
|
| 113 |
"ringStop": 1,
|
| 114 |
"ringStaging": 1,
|
| 115 |
"disableCommandBatching": false
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 116 |
}
|
| 117 |
},
|
| 118 |
"execution": {
|
|
@@ -140,7 +163,10 @@
|
|
| 140 |
"attn_decode": {
|
| 141 |
"kernel": "attention_decode_online_f16kv.wgsl",
|
| 142 |
"entry": "main",
|
| 143 |
-
"digest": "sha256:
|
|
|
|
|
|
|
|
|
|
| 144 |
},
|
| 145 |
"residual": {
|
| 146 |
"kernel": "residual.wgsl",
|
|
@@ -168,12 +194,18 @@
|
|
| 168 |
"attn_small": {
|
| 169 |
"kernel": "attention_small_f16kv.wgsl",
|
| 170 |
"entry": "main",
|
| 171 |
-
"digest": "sha256:7e3acc24b8b45294d18052c759319fe8a202d4ebfc4b653d4b904b245ae7e5c9"
|
|
|
|
|
|
|
|
|
|
| 172 |
},
|
| 173 |
"attn_head256": {
|
| 174 |
"kernel": "attention_head256_f16kv.wgsl",
|
| 175 |
"entry": "main",
|
| 176 |
-
"digest": "sha256:4ecdc079e322a770350414244cb383d72ae7ffa47ed620c64ff425c93413e97a"
|
|
|
|
|
|
|
|
|
|
| 177 |
},
|
| 178 |
"lm_head_gemv": {
|
| 179 |
"kernel": "matmul_gemv_subgroup.wgsl",
|
|
@@ -4375,7 +4407,7 @@
|
|
| 4375 |
"convertedAt": "2026-02-26T20:19:47.896Z",
|
| 4376 |
"hasTokenizer": true,
|
| 4377 |
"manifestRefresh": {
|
| 4378 |
-
"at": "2026-04-
|
| 4379 |
"source": "local-hash-resync",
|
| 4380 |
"config": "gemma-3-1b-it-q4k-ehf16-af32.json"
|
| 4381 |
}
|
|
|
|
| 35 |
"valueNorm": false
|
| 36 |
},
|
| 37 |
"normalization": {
|
| 38 |
+
"rmsNormEps": 0.000001,
|
| 39 |
"rmsNormWeightOffset": true,
|
| 40 |
"postAttentionNorm": true,
|
| 41 |
"preFeedforwardNorm": true,
|
|
|
|
| 113 |
"ringStop": 1,
|
| 114 |
"ringStaging": 1,
|
| 115 |
"disableCommandBatching": false
|
| 116 |
+
},
|
| 117 |
+
"perLayerInputs": {
|
| 118 |
+
"materialization": "auto",
|
| 119 |
+
"rowCache": {
|
| 120 |
+
"mode": "lru",
|
| 121 |
+
"maxRows": 256,
|
| 122 |
+
"maxBytes": 134217728,
|
| 123 |
+
"decodedDtype": "f32"
|
| 124 |
+
},
|
| 125 |
+
"prefetch": {
|
| 126 |
+
"mode": "next_token",
|
| 127 |
+
"rowsAhead": 1
|
| 128 |
+
},
|
| 129 |
+
"gpuUpload": {
|
| 130 |
+
"mode": "per_step_slices",
|
| 131 |
+
"stagingRows": 2
|
| 132 |
+
},
|
| 133 |
+
"hotCache": {
|
| 134 |
+
"mode": "prepared_tokens",
|
| 135 |
+
"maxTokens": 1024,
|
| 136 |
+
"maxBytes": 134217728,
|
| 137 |
+
"outputDtype": "f32"
|
| 138 |
+
}
|
| 139 |
}
|
| 140 |
},
|
| 141 |
"execution": {
|
|
|
|
| 163 |
"attn_decode": {
|
| 164 |
"kernel": "attention_decode_online_f16kv.wgsl",
|
| 165 |
"entry": "main",
|
| 166 |
+
"digest": "sha256:beee4722aa6281f58b74e1d055e192017b5555c189a58dd13dd158fe1511a7b8",
|
| 167 |
+
"precision": {
|
| 168 |
+
"kvDtype": "f16"
|
| 169 |
+
}
|
| 170 |
},
|
| 171 |
"residual": {
|
| 172 |
"kernel": "residual.wgsl",
|
|
|
|
| 194 |
"attn_small": {
|
| 195 |
"kernel": "attention_small_f16kv.wgsl",
|
| 196 |
"entry": "main",
|
| 197 |
+
"digest": "sha256:7e3acc24b8b45294d18052c759319fe8a202d4ebfc4b653d4b904b245ae7e5c9",
|
| 198 |
+
"precision": {
|
| 199 |
+
"kvDtype": "f16"
|
| 200 |
+
}
|
| 201 |
},
|
| 202 |
"attn_head256": {
|
| 203 |
"kernel": "attention_head256_f16kv.wgsl",
|
| 204 |
"entry": "main",
|
| 205 |
+
"digest": "sha256:4ecdc079e322a770350414244cb383d72ae7ffa47ed620c64ff425c93413e97a",
|
| 206 |
+
"precision": {
|
| 207 |
+
"kvDtype": "f16"
|
| 208 |
+
}
|
| 209 |
},
|
| 210 |
"lm_head_gemv": {
|
| 211 |
"kernel": "matmul_gemv_subgroup.wgsl",
|
|
|
|
| 4407 |
"convertedAt": "2026-02-26T20:19:47.896Z",
|
| 4408 |
"hasTokenizer": true,
|
| 4409 |
"manifestRefresh": {
|
| 4410 |
+
"at": "2026-04-09T01:07:09.823Z",
|
| 4411 |
"source": "local-hash-resync",
|
| 4412 |
"config": "gemma-3-1b-it-q4k-ehf16-af32.json"
|
| 4413 |
}
|