@bendyline/gilde 0.1.3 → 0.1.5

package/data/chat-models/ge/gemma4-31b-q4/manifest.json

@@ -3,21 +3,30 @@
   "kind": "chat-model",
   "id": "gemma4-31b-q4",
   "name": "Gemma 4 (31B)",
-  "description": "Google's flagship Gemma 4 at 30.7B parameters. Multimodal (text + image + video), 256K context, configurable thinking mode, native function-calling, 140+ languages. Frontier-tier instruction-following on device \u2014 needs 32GB+ of usable memory.",
-  "tags": ["google", "multimodal", "vision", "video", "tools", "reasoning", "large", "qat"],
+  "description": "Google's flagship Gemma 4 at 30.7B parameters. Multimodal (text + image + video), 256K context, configurable thinking mode, native function-calling, 140+ languages. Frontier-tier instruction-following on device — needs 32GB+ of usable memory.",
+  "tags": [
+    "google",
+    "multimodal",
+    "vision",
+    "video",
+    "tools",
+    "reasoning",
+    "large",
+    "qat"
+  ],
   "category": "general",
   "maintainer": {
     "name": "Google",
     "url": "https://huggingface.co/google/gemma-4-31B-it"
   },
-  "version": "1.2.0",
-  "updatedAt": "2026-06-06T00:00:00Z",
+  "version": "1.2.1",
+  "updatedAt": "2026-07-28T00:00:00Z",
   "license": "Apache-2.0",
   "licenseClass": "open",
   "licenseShortName": "Apache 2.0",
   "recoScore": 15,
   "parameterSize": "30.7B",
-  "approxSizeBytes": 17287668064,
+  "approxSizeBytes": 17287670048,
   "supportsTools": true,
   "contextWindow": 256000,
   "upstream": "https://huggingface.co/google/gemma-4-31B-it",
@@ -123,6 +132,14 @@
           "maxTokens": 8192
         }
       }
+    },
+    "engine": {
+      "llamaCpp": {
+        "spec": {
+          "mtp": true,
+          "nMax": 4
+        }
+      }
     }
   },
   "ollama": {
@@ -130,18 +147,23 @@
   },
   "llamaCpp": {
     "huggingfaceRepo": "unsloth/gemma-4-31B-it-qat-GGUF",
-    "revision": "840527d199d2988bdf3656cd8d373e82ea50cb1e",
+    "revision": "43cc1aeb31adf47ec06a854507ce552cd9862e6f",
     "filename": "gemma-4-31B-it-qat-UD-Q4_K_XL.gguf",
-    "sha256": "9188a71055550f1e60b875d02b7abb63625ac11b4a6f148d6b22b3b28ba3d335",
-    "approxSizeBytes": 17287668064,
+    "sha256": "00b5a7c497f0c8934033088c10a7fa9a4c015e46ee6d89e9c6890650ba5d0e71",
+    "approxSizeBytes": 17287670048,
     "quantization": "UD-Q4_K_XL",
-    "residentBytes": 21030000000
+    "residentBytes": 21081151219,
+    "draftModel": {
+      "filename": "MTP/mtp-gemma-4-31B-it-Q4_0.gguf",
+      "sha256": "3a5e99fd8d0b23afb1fccd1ee0c9ebd1f571d00399c2dae2292d217feeec0f6b",
+      "sizeBytes": 279955968
+    }
   },
   "mlx": {
     "huggingfaceRepo": "mlx-community/gemma-4-31b-it-4bit",
-    "revision": "0d17175ead577037577f24963de2f0f5fafb72a5",
+    "revision": "696d436c404745a59f30e4939a658162b0a9e57f",
     "quantization": "4bit",
-    "approxSizeBytes": 18444417871,
+    "approxSizeBytes": 18444419448,
     "files": [
       {
         "name": "chat_template.jinja",
@@ -150,8 +172,8 @@
       },
       {
         "name": "config.json",
-        "sha256": "f24fd0f1114e77bc4b3ffb539292702aa63806eed63669d2ada9d549705f51e2",
-        "sizeBytes": 5647
+        "sha256": "a01be3d45e0eb70ae89cef6d3e823dd44de6ec4dfc41b91dc357f97e457915c0",
+        "sizeBytes": 6046
       },
       {
         "name": "generation_config.json",
@@ -160,23 +182,23 @@
       },
       {
         "name": "model-00001-of-00004.safetensors",
-        "sha256": "ea3de2445014c29cdb149b3d291b867a5814bd33b817ac1cb2b46532811bdd79",
-        "sizeBytes": 5366617542
+        "sha256": "988e2b1fd41d93b62b8c432f52f632c43b8cb7f86df4b957db36a3cc0dab40ca",
+        "sizeBytes": 5366617512
       },
       {
         "name": "model-00002-of-00004.safetensors",
-        "sha256": "d3d5b9517c1c153fb3aa57435c14b2fc7ff026b3a85b0cb692458f49d8831079",
-        "sizeBytes": 5361642569
+        "sha256": "a496a96fbd39cd11a9871f91d026013d91069dcac15f89ca861b93976f3857cf",
+        "sizeBytes": 5361642573
       },
       {
         "name": "model-00003-of-00004.safetensors",
-        "sha256": "6c133e72192f9ba547ec03c3b2380ceabe572ff345dc92a91c76caae0f1b609e",
-        "sizeBytes": 5367276098
+        "sha256": "afa555ff0e1bc458c5b08aeef1f4499dce63e2bfb5d3a2aac716e47c0a5672c1",
+        "sizeBytes": 5367276094
       },
       {
         "name": "model-00004-of-00004.safetensors",
-        "sha256": "3509efbc779aa901c51ec9d398170d3c60e9c8d1b51b8fee64b27680a3d55aef",
-        "sizeBytes": 2316480623
+        "sha256": "b55567ae065357d047daca926e45619ec0edef994f165356111c95d4d1021587",
+        "sizeBytes": 2316480497
       },
       {
         "name": "model.safetensors.index.json",
@@ -185,8 +207,8 @@
       },
       {
         "name": "processor_config.json",
-        "sha256": "50c9cf588f1bda1c93d92ec69b03011bf101cc6867c6415fe5f07f1c87e49e72",
-        "sizeBytes": 627
+        "sha256": "de3e580aebdc98272d4c4547daffe6525fcbae18a83a0e0bcf0d7444d4ee6f37",
+        "sizeBytes": 1316
       },
       {
         "name": "tokenizer.json",
@@ -195,8 +217,8 @@
       },
       {
         "name": "tokenizer_config.json",
-        "sha256": "90c3a3ba5bf53818383a58e1a776cbcacd2a038d4812eaa373e1522f2d06f3df",
-        "sizeBytes": 2095
+        "sha256": "080d9e1aff284e2f6043889cd05367966f7c7b80e025fbc0b06745e218158656",
+        "sizeBytes": 2740
       }
     ],
     "residentBytes": 23977741909

package/data/chat-models/ge/gemma4-31b-q4/versions/1.2.1/manifest.json

@@ -0,0 +1,87 @@
+{
+  "schemaVersion": 1,
+  "version": "1.2.1",
+  "releasedAt": "2026-07-28T00:00:00Z",
+  "approxSizeBytes": 17287670048,
+  "ollama": {
+    "tag": "gemma4:31b-it-qat"
+  },
+  "llamaCpp": {
+    "huggingfaceRepo": "unsloth/gemma-4-31B-it-qat-GGUF",
+    "revision": "43cc1aeb31adf47ec06a854507ce552cd9862e6f",
+    "filename": "gemma-4-31B-it-qat-UD-Q4_K_XL.gguf",
+    "sha256": "00b5a7c497f0c8934033088c10a7fa9a4c015e46ee6d89e9c6890650ba5d0e71",
+    "approxSizeBytes": 17287670048,
+    "quantization": "UD-Q4_K_XL",
+    "residentBytes": 21081151219,
+    "draftModel": {
+      "filename": "MTP/mtp-gemma-4-31B-it-Q4_0.gguf",
+      "sha256": "3a5e99fd8d0b23afb1fccd1ee0c9ebd1f571d00399c2dae2292d217feeec0f6b",
+      "sizeBytes": 279955968
+    }
+  },
+  "mlx": {
+    "huggingfaceRepo": "mlx-community/gemma-4-31b-it-4bit",
+    "revision": "696d436c404745a59f30e4939a658162b0a9e57f",
+    "quantization": "4bit",
+    "approxSizeBytes": 18444419448,
+    "files": [
+      {
+        "name": "chat_template.jinja",
+        "sha256": "36e3a42e5cf14cd0020e72d92e1fdd9970f59b82170e421f0cbe1bb42bead3f0",
+        "sizeBytes": 17466
+      },
+      {
+        "name": "config.json",
+        "sha256": "a01be3d45e0eb70ae89cef6d3e823dd44de6ec4dfc41b91dc357f97e457915c0",
+        "sizeBytes": 6046
+      },
+      {
+        "name": "generation_config.json",
+        "sha256": "d4226bbe3117d2d253ba4609720ba82c6c4ce4627a9a6ae05387c78983ac03de",
+        "sizeBytes": 208
+      },
+      {
+        "name": "model-00001-of-00004.safetensors",
+        "sha256": "988e2b1fd41d93b62b8c432f52f632c43b8cb7f86df4b957db36a3cc0dab40ca",
+        "sizeBytes": 5366617512
+      },
+      {
+        "name": "model-00002-of-00004.safetensors",
+        "sha256": "a496a96fbd39cd11a9871f91d026013d91069dcac15f89ca861b93976f3857cf",
+        "sizeBytes": 5361642573
+      },
+      {
+        "name": "model-00003-of-00004.safetensors",
+        "sha256": "afa555ff0e1bc458c5b08aeef1f4499dce63e2bfb5d3a2aac716e47c0a5672c1",
+        "sizeBytes": 5367276094
+      },
+      {
+        "name": "model-00004-of-00004.safetensors",
+        "sha256": "b55567ae065357d047daca926e45619ec0edef994f165356111c95d4d1021587",
+        "sizeBytes": 2316480497
+      },
+      {
+        "name": "model.safetensors.index.json",
+        "sha256": "a9ca8373aacc37ab9d7f16ff4bbb392c055e1b51b0fec6258cbfc578056c5fef",
+        "sizeBytes": 205370
+      },
+      {
+        "name": "processor_config.json",
+        "sha256": "de3e580aebdc98272d4c4547daffe6525fcbae18a83a0e0bcf0d7444d4ee6f37",
+        "sizeBytes": 1316
+      },
+      {
+        "name": "tokenizer.json",
+        "sha256": "cc8d3a0ce36466ccc1278bf987df5f71db1719b9ca6b4118264f45cb627bfe0f",
+        "sizeBytes": 32169626
+      },
+      {
+        "name": "tokenizer_config.json",
+        "sha256": "080d9e1aff284e2f6043889cd05367966f7c7b80e025fbc0b06745e218158656",
+        "sizeBytes": 2740
+      }
+    ],
+    "residentBytes": 23977741909
+  }
+}

package/data/chat-models/ge/gemma4-e2b-q8/manifest.json

@@ -17,14 +17,14 @@
     "name": "Google",
     "url": "https://huggingface.co/google/gemma-4-E2B-it"
   },
-  "version": "1.1.1",
-  "updatedAt": "2026-04-26T00:00:00Z",
+  "version": "1.1.2",
+  "updatedAt": "2026-07-28T00:00:00Z",
   "license": "Apache-2.0",
   "licenseClass": "open",
   "licenseShortName": "Apache 2.0",
   "recoScore": 15,
   "parameterSize": "2.3B",
-  "approxSizeBytes": 4967494592,
+  "approxSizeBytes": 4967497152,
   "supportsTools": true,
   "contextWindow": 128000,
   "upstream": "https://huggingface.co/google/gemma-4-E2B-it",
@@ -55,67 +55,6 @@
     "prompt.derive-by-execution",
     "tools.mlx-grammar"
   ],
-  "ollama": {
-    "tag": "gemma4:e2b"
-  },
-  "llamaCpp": {
-    "huggingfaceRepo": "ggml-org/gemma-4-E2B-it-GGUF",
-    "revision": "a1dac71d3ab220618f5a7573a52acdc4baf3ae3b",
-    "filename": "gemma-4-E2B-it-Q8_0.gguf",
-    "sha256": "e049411c01fb7a81161768c52e38828970e55a64e22738957adcbe51d20f1c8e",
-    "approxSizeBytes": 4967494592,
-    "quantization": "Q8_0",
-    "residentBytes": 5960993510
-  },
-  "mlx": {
-    "huggingfaceRepo": "mlx-community/gemma-4-e2b-it-4bit",
-    "revision": "2c3e507453b4f218d05fe3cc97bea5c5a654257e",
-    "quantization": "4bit",
-    "approxSizeBytes": 3613527369,
-    "files": [
-      {
-        "name": "chat_template.jinja",
-        "sha256": "2f1b4d75d067bae3fe44e676721c7f077d243bc007156cb9c2f8b5836613d082",
-        "sizeBytes": 17336
-      },
-      {
-        "name": "config.json",
-        "sha256": "6d12c87861fff3871d3a745011b0d852be6513f3ce594ae1e8d643dae9d3b9a8",
-        "sizeBytes": 5996
-      },
-      {
-        "name": "generation_config.json",
-        "sha256": "d4226bbe3117d2d253ba4609720ba82c6c4ce4627a9a6ae05387c78983ac03de",
-        "sizeBytes": 208
-      },
-      {
-        "name": "model.safetensors",
-        "sha256": "e9bea0584546fafb5ff83a1132a6c4662a8498cc6a5bcda52fc6ca562b7bafab",
-        "sizeBytes": 3581101896
-      },
-      {
-        "name": "model.safetensors.index.json",
-        "sha256": "a8aa7359c747a0d59368dbff9a1029da86bda139ccc0ae1f1e938db75de7d5ce",
-        "sizeBytes": 230329
-      },
-      {
-        "name": "processor_config.json",
-        "sha256": "1bd0d00776284f369c1eff5fb631e865dfcdca861e0b7d60dbef27fcf37436a8",
-        "sizeBytes": 902
-      },
-      {
-        "name": "tokenizer.json",
-        "sha256": "cc8d3a0ce36466ccc1278bf987df5f71db1719b9ca6b4118264f45cb627bfe0f",
-        "sizeBytes": 32169626
-      },
-      {
-        "name": "tokenizer_config.json",
-        "sha256": "90c3a3ba5bf53818383a58e1a776cbcacd2a038d4812eaa373e1522f2d06f3df",
-        "sizeBytes": 2095
-      }
-    ],
-    "residentBytes": 4697585580
-  },
   "tuning": {
     "sampling": {
       "temperature": 1,
@@ -194,6 +133,79 @@
           "repetitionContext": 20
         }
       }
+    },
+    "engine": {
+      "llamaCpp": {
+        "spec": {
+          "mtp": true,
+          "nMax": 4
+        }
+      }
+    }
+  },
+  "ollama": {
+    "tag": "gemma4:e2b"
+  },
+  "llamaCpp": {
+    "huggingfaceRepo": "ggml-org/gemma-4-E2B-it-GGUF",
+    "revision": "b4243c156154b6dca9324415f8c7ccc098b4aed1",
+    "filename": "gemma-4-E2B-it-Q8_0.gguf",
+    "sha256": "996d08777aadc6bfd3c7375ef70ba25a0f55240075860754fdb18d6d860aa63a",
+    "approxSizeBytes": 4967497152,
+    "quantization": "Q8_0",
+    "residentBytes": 6078377818,
+    "draftModel": {
+      "filename": "mtp-gemma-4-E2B-it-Q8_0.gguf",
+      "sha256": "c4fba8d43b40c9fab8c3db15ca6ef00fd28192208753f1f038269c176437068a",
+      "sizeBytes": 97817696
     }
+  },
+  "mlx": {
+    "huggingfaceRepo": "mlx-community/gemma-4-e2b-it-4bit",
+    "revision": "238767527555cb75a05732a84dff5d6ba0dd6809",
+    "quantization": "4bit",
+    "approxSizeBytes": 3583086498,
+    "files": [
+      {
+        "name": "chat_template.jinja",
+        "sha256": "2f1b4d75d067bae3fe44e676721c7f077d243bc007156cb9c2f8b5836613d082",
+        "sizeBytes": 17336
+      },
+      {
+        "name": "config.json",
+        "sha256": "6397cb6eca41b911d1dcab74e17941351057bd759284052a2331918ff6f9246c",
+        "sizeBytes": 6395
+      },
+      {
+        "name": "generation_config.json",
+        "sha256": "d4226bbe3117d2d253ba4609720ba82c6c4ce4627a9a6ae05387c78983ac03de",
+        "sizeBytes": 208
+      },
+      {
+        "name": "model.safetensors",
+        "sha256": "038e39a37a7667373d2c3991375446b10c96ae1d717a68674870343db376b76e",
+        "sizeBytes": 3550670554
+      },
+      {
+        "name": "model.safetensors.index.json",
+        "sha256": "edb157dbf495e23f37377af4a628a9ad13c4ee7937f93ccb36ec9e9a19940f16",
+        "sizeBytes": 218323
+      },
+      {
+        "name": "processor_config.json",
+        "sha256": "de3e580aebdc98272d4c4547daffe6525fcbae18a83a0e0bcf0d7444d4ee6f37",
+        "sizeBytes": 1316
+      },
+      {
+        "name": "tokenizer.json",
+        "sha256": "cc8d3a0ce36466ccc1278bf987df5f71db1719b9ca6b4118264f45cb627bfe0f",
+        "sizeBytes": 32169626
+      },
+      {
+        "name": "tokenizer_config.json",
+        "sha256": "080d9e1aff284e2f6043889cd05367966f7c7b80e025fbc0b06745e218158656",
+        "sizeBytes": 2740
+      }
+    ]
   }
 }

package/data/chat-models/ge/gemma4-e2b-q8/versions/1.1.2/manifest.json

@@ -0,0 +1,71 @@
+{
+  "schemaVersion": 1,
+  "version": "1.1.2",
+  "releasedAt": "2026-07-28T00:00:00Z",
+  "approxSizeBytes": 4967497152,
+  "ollama": {
+    "tag": "gemma4:e2b"
+  },
+  "llamaCpp": {
+    "huggingfaceRepo": "ggml-org/gemma-4-E2B-it-GGUF",
+    "revision": "b4243c156154b6dca9324415f8c7ccc098b4aed1",
+    "filename": "gemma-4-E2B-it-Q8_0.gguf",
+    "sha256": "996d08777aadc6bfd3c7375ef70ba25a0f55240075860754fdb18d6d860aa63a",
+    "approxSizeBytes": 4967497152,
+    "quantization": "Q8_0",
+    "residentBytes": 6078377818,
+    "draftModel": {
+      "filename": "mtp-gemma-4-E2B-it-Q8_0.gguf",
+      "sha256": "c4fba8d43b40c9fab8c3db15ca6ef00fd28192208753f1f038269c176437068a",
+      "sizeBytes": 97817696
+    }
+  },
+  "mlx": {
+    "huggingfaceRepo": "mlx-community/gemma-4-e2b-it-4bit",
+    "revision": "238767527555cb75a05732a84dff5d6ba0dd6809",
+    "quantization": "4bit",
+    "approxSizeBytes": 3583086498,
+    "files": [
+      {
+        "name": "chat_template.jinja",
+        "sha256": "2f1b4d75d067bae3fe44e676721c7f077d243bc007156cb9c2f8b5836613d082",
+        "sizeBytes": 17336
+      },
+      {
+        "name": "config.json",
+        "sha256": "6397cb6eca41b911d1dcab74e17941351057bd759284052a2331918ff6f9246c",
+        "sizeBytes": 6395
+      },
+      {
+        "name": "generation_config.json",
+        "sha256": "d4226bbe3117d2d253ba4609720ba82c6c4ce4627a9a6ae05387c78983ac03de",
+        "sizeBytes": 208
+      },
+      {
+        "name": "model.safetensors",
+        "sha256": "038e39a37a7667373d2c3991375446b10c96ae1d717a68674870343db376b76e",
+        "sizeBytes": 3550670554
+      },
+      {
+        "name": "model.safetensors.index.json",
+        "sha256": "edb157dbf495e23f37377af4a628a9ad13c4ee7937f93ccb36ec9e9a19940f16",
+        "sizeBytes": 218323
+      },
+      {
+        "name": "processor_config.json",
+        "sha256": "de3e580aebdc98272d4c4547daffe6525fcbae18a83a0e0bcf0d7444d4ee6f37",
+        "sizeBytes": 1316
+      },
+      {
+        "name": "tokenizer.json",
+        "sha256": "cc8d3a0ce36466ccc1278bf987df5f71db1719b9ca6b4118264f45cb627bfe0f",
+        "sizeBytes": 32169626
+      },
+      {
+        "name": "tokenizer_config.json",
+        "sha256": "080d9e1aff284e2f6043889cd05367966f7c7b80e025fbc0b06745e218158656",
+        "sizeBytes": 2740
+      }
+    ]
+  }
+}

package/data/chat-models/ge/gemma4-e4b-q8/manifest.json

@@ -16,14 +16,14 @@
     "name": "Google",
     "url": "https://huggingface.co/google/gemma-4-E4B-it"
   },
-  "version": "1.1.1",
-  "updatedAt": "2026-06-07T00:00:00Z",
+  "version": "1.1.2",
+  "updatedAt": "2026-07-28T00:00:00Z",
   "license": "Apache-2.0",
   "licenseClass": "open",
   "licenseShortName": "Apache 2.0",
   "recoScore": 15,
   "parameterSize": "8B",
-  "approxSizeBytes": 8031240160,
+  "approxSizeBytes": 8031242688,
   "supportsTools": true,
   "contextWindow": 128000,
   "upstream": "https://huggingface.co/google/gemma-4-E4B-it",
@@ -50,72 +50,6 @@
     "prompt.derive-by-execution",
     "tools.mlx-grammar"
   ],
-  "ollama": {
-    "tag": "gemma4:e4b"
-  },
-  "llamaCpp": {
-    "huggingfaceRepo": "ggml-org/gemma-4-E4B-it-GGUF",
-    "revision": "2714b5519c6c3516b1000e7c5e1eba998dfe1fe8",
-    "filename": "gemma-4-E4B-it-Q8_0.gguf",
-    "sha256": "fb8f0c032de00b18c710824af3c7e5777c71e5fb60b13f13575f0a9e92ddecd0",
-    "approxSizeBytes": 8031240160,
-    "quantization": "Q8_0",
-    "residentBytes": 9637488192
-  },
-  "mlx": {
-    "huggingfaceRepo": "mlx-community/gemma-4-E4B-it-qat-4bit",
-    "revision": "0f35c6f6d386f7f74e628bd7c6526ce531212300",
-    "files": [
-      {
-        "name": "chat_template.jinja",
-        "sha256": "33204f1acb5bd0002713e16a593847f24ceeafe711ed88bda2a352dc996a3373",
-        "sizeBytes": 16804
-      },
-      {
-        "name": "config.json",
-        "sha256": "7af8ec026a79318244e56c7df64c62a36cfa7ce361001112db9d919b5e18e9e2",
-        "sizeBytes": 36485
-      },
-      {
-        "name": "generation_config.json",
-        "sha256": "b69207f9be617e982d13cc273cce6fd88c98dda99a4bdc5e2d52ffe0a0d9f0a9",
-        "sizeBytes": 203
-      },
-      {
-        "name": "model-00001-of-00002.safetensors",
-        "sha256": "6c8525b3cecdc9847fc5606ea4ad9445c1496072d99812283eea843a06d5f58b",
-        "sizeBytes": 4249502053
-      },
-      {
-        "name": "model-00002-of-00002.safetensors",
-        "sha256": "5c5715153fbb71079014d91e551229e0da09494d98ff9dcec0977ac8b4d58d8c",
-        "sizeBytes": 2548805689
-      },
-      {
-        "name": "model.safetensors.index.json",
-        "sha256": "128114effccfb1b772c80517b3cbcbc2a7813e28eb8c71dc56e114450c9e6429",
-        "sizeBytes": 282511
-      },
-      {
-        "name": "processor_config.json",
-        "sha256": "1bd0d00776284f369c1eff5fb631e865dfcdca861e0b7d60dbef27fcf37436a8",
-        "sizeBytes": 902
-      },
-      {
-        "name": "tokenizer.json",
-        "sha256": "cc8d3a0ce36466ccc1278bf987df5f71db1719b9ca6b4118264f45cb627bfe0f",
-        "sizeBytes": 32169626
-      },
-      {
-        "name": "tokenizer_config.json",
-        "sha256": "080d9e1aff284e2f6043889cd05367966f7c7b80e025fbc0b06745e218158656",
-        "sizeBytes": 2740
-      }
-    ],
-    "approxSizeBytes": 6830817013,
-    "quantization": "4bit-qat",
-    "disabledReason": "mlx-vlm cannot load Gemma-4 E4B (elastic MatFormer + audio tower) — fails on every published build across mlx-vlm 0.6.1-0.6.4."
-  },
   "tuning": {
     "sampling": {
       "temperature": 1,
@@ -132,9 +66,8 @@
     "engine": {
       "llamaCpp": {
         "spec": {
-          "type": "draft-simple",
-          "draftModelId": "gemma4-e2b-q8",
-          "nMax": 4
+          "nMax": 4,
+          "mtp": true
         }
       }
     },
@@ -182,7 +115,7 @@
       },
       "creative": {
         "sampling": {
-          "temperature": 1.0,
+          "temperature": 1,
           "topP": 0.95,
           "topK": 80,
           "repetitionPenalty": 1.1,
@@ -191,5 +124,70 @@
         }
       }
     }
+  },
+  "ollama": {
+    "tag": "gemma4:e4b"
+  },
+  "llamaCpp": {
+    "huggingfaceRepo": "ggml-org/gemma-4-E4B-it-GGUF",
+    "revision": "b8093469224f83f5c38f691eb906c380e9e63114",
+    "filename": "gemma-4-E4B-it-Q8_0.gguf",
+    "sha256": "34be82b17b4942d389b9b527170c4b058027abdd32531fda063d3d97dd8ce80a",
+    "approxSizeBytes": 8031242688,
+    "quantization": "Q8_0",
+    "residentBytes": 9755875162,
+    "draftModel": {
+      "filename": "mtp-gemma-4-E4B-it-Q8_0.gguf",
+      "sha256": "f38ae62962657c7a6303c49bbb147e9ae23634e911cfa532fac0818c2e18b665",
+      "sizeBytes": 98653280
+    }
+  },
+  "mlx": {
+    "huggingfaceRepo": "mlx-community/gemma-4-e4b-it-4bit",
+    "revision": "475b9088d29754a3379866cf5aeb6b41acd313c2",
+    "quantization": "4bit",
+    "approxSizeBytes": 5179239349,
+    "files": [
+      {
+        "name": "chat_template.jinja",
+        "sha256": "2f1b4d75d067bae3fe44e676721c7f077d243bc007156cb9c2f8b5836613d082",
+        "sizeBytes": 17336
+      },
+      {
+        "name": "config.json",
+        "sha256": "780ccb3a514a5f1ced161d383f948fc22eca9b84b752ca19494f625bd9bad7a6",
+        "sizeBytes": 6628
+      },
+      {
+        "name": "generation_config.json",
+        "sha256": "d4226bbe3117d2d253ba4609720ba82c6c4ce4627a9a6ae05387c78983ac03de",
+        "sizeBytes": 208
+      },
+      {
+        "name": "model.safetensors",
+        "sha256": "932b8271fc3fe65adcc78b96c10c6268bbfb13e8f67d1358727c0d6ee97e1eff",
+        "sizeBytes": 5146800534
+      },
+      {
+        "name": "model.safetensors.index.json",
+        "sha256": "f8accac59ee7efe87e0c298c854610b262c3cadd477407503147c71209ff0093",
+        "sizeBytes": 240961
+      },
+      {
+        "name": "processor_config.json",
+        "sha256": "de3e580aebdc98272d4c4547daffe6525fcbae18a83a0e0bcf0d7444d4ee6f37",
+        "sizeBytes": 1316
+      },
+      {
+        "name": "tokenizer.json",
+        "sha256": "cc8d3a0ce36466ccc1278bf987df5f71db1719b9ca6b4118264f45cb627bfe0f",
+        "sizeBytes": 32169626
+      },
+      {
+        "name": "tokenizer_config.json",
+        "sha256": "080d9e1aff284e2f6043889cd05367966f7c7b80e025fbc0b06745e218158656",
+        "sizeBytes": 2740
+      }
+    ]
   }
 }