@bendyline/gezel-inference 0.1.0 → 1.0.1

package/dist/catalog/chat-models/lf/lfm2.5-24b-a2b-q4/manifest.json

@@ -0,0 +1,122 @@
+{
+  "schemaVersion": 1,
+  "kind": "chat-model",
+  "id": "lfm2.5-24b-a2b-q4",
+  "name": "LFM2.5 (24B-A2B, Q4)",
+  "description": "Liquid AI's largest LFM2.5, a mixture-of-experts model that holds 24B parameters but uses only about 2B for each word it writes, so it stays quick for its size. Built for agent work: it always thinks before it answers, calls tools natively, and has a 128K context window. Pair it with search or your documents for fact-heavy questions.",
+  "tags": [
+    "liquid-ai",
+    "agentic",
+    "reasoning",
+    "tools",
+    "moe"
+  ],
+  "category": "general",
+  "maintainer": {
+    "name": "Liquid AI",
+    "url": "https://huggingface.co/LiquidAI/LFM2.5-24B-A2B"
+  },
+  "version": "1.0.0",
+  "updatedAt": "2026-10-14T00:00:00Z",
+  "license": "LFM Open License v1.0",
+  "licenseClass": "custom-restricted",
+  "licenseShortName": "LFM Open 1.0",
+  "parameterSize": "24B",
+  "approxSizeBytes": 13578190880,
+  "supportsTools": true,
+  "contextWindow": 128000,
+  "upstream": "https://huggingface.co/LiquidAI/LFM2.5-24B-A2B",
+  "style": {
+    "family": "other",
+    "reasoningFormat": "think",
+    "toolCallFormat": "function-call"
+  },
+  "behaviors": [
+    "reasoning.strip-think-tags",
+    "prompt.private-reasoning-guidance",
+    "prompt.tool-cookbook-full",
+    "fabrication.detect-past-tense-no-tools",
+    "mcp.compact-tool-schemas",
+    "mcp.relax-required-fields",
+    "mcp.default-missing-fields",
+    {
+      "id": "turn.continuation-budget",
+      "config": {
+        "count": 4
+      }
+    },
+    "turn.preamble-folding",
+    "turn.ramble-detection",
+    "prompt.prefer-writefile-edits",
+    "prompt.derive-by-execution",
+    "prompt.retrieval-first",
+    "prompt.workspace-gestalt",
+    {
+      "id": "prompt.native-tool-call-format",
+      "config": {
+        "example": "<|tool_call_start|>[write_file(path='notes.md', content='hello')]<|tool_call_end|>"
+      }
+    }
+  ],
+  "tuning": {
+    "sampling": {
+      "temperature": 0.1,
+      "topK": 50,
+      "repetitionPenalty": 1.05
+    },
+    "reasoning": {
+      "enableThinking": true,
+      "thinkingBudget": 4096
+    },
+    "profiles": {
+      "thinking-general": {
+        "sampling": {
+          "temperature": 0.1,
+          "topP": 0.95,
+          "topK": 50,
+          "repetitionPenalty": 1.05
+        }
+      },
+      "thinking-coding": {
+        "sampling": {
+          "temperature": 0.1,
+          "topP": 0.9,
+          "topK": 50,
+          "repetitionPenalty": 1.05
+        }
+      },
+      "thinking-precise": {
+        "sampling": {
+          "temperature": 0.05,
+          "topP": 0.9,
+          "topK": 20,
+          "repetitionPenalty": 1.05
+        }
+      },
+      "instruct": {
+        "sampling": {
+          "temperature": 0.1,
+          "topP": 0.95,
+          "topK": 50,
+          "repetitionPenalty": 1.05
+        }
+      },
+      "creative": {
+        "sampling": {
+          "temperature": 0.3,
+          "topP": 0.95,
+          "topK": 50,
+          "repetitionPenalty": 1.05
+        }
+      }
+    }
+  },
+  "llamaCpp": {
+    "huggingfaceRepo": "LiquidAI/LFM2.5-24B-A2B-GGUF",
+    "revision": "0912d33141f77e10705345a6732bfd83f0252fdc",
+    "filename": "LFM2.5-24B-A2B-QAD-Q4_0.gguf",
+    "sha256": "d1c03f73ecaf61017d70419eec9e5dc0b05d50ae26c23d522829b934fbabf8a5",
+    "approxSizeBytes": 13578190880,
+    "quantization": "Q4_0"
+  }
+}

package/dist/catalog/chat-models/lf/lfm2.5-24b-a2b-q4/versions/1.0.0/manifest.json

@@ -0,0 +1,15 @@
+{
+  "schemaVersion": 1,
+  "version": "1.0.0",
+  "releasedAt": "2026-10-14T00:00:00Z",
+  "minGezelVersion": "1.26287",
+  "approxSizeBytes": 13578190880,
+  "llamaCpp": {
+    "huggingfaceRepo": "LiquidAI/LFM2.5-24B-A2B-GGUF",
+    "revision": "0912d33141f77e10705345a6732bfd83f0252fdc",
+    "filename": "LFM2.5-24B-A2B-QAD-Q4_0.gguf",
+    "sha256": "d1c03f73ecaf61017d70419eec9e5dc0b05d50ae26c23d522829b934fbabf8a5",
+    "approxSizeBytes": 13578190880,
+    "quantization": "Q4_0"
+  }
+}

package/dist/catalog/chat-models/mi/minicpm-v4.7-35b-a3b-q4/manifest.json

@@ -0,0 +1,185 @@
+{
+  "schemaVersion": 1,
+  "kind": "chat-model",
+  "id": "minicpm-v4.7-35b-a3b-q4",
+  "name": "MiniCPM-V 4.7 (35B-A3B, Q4)",
+  "description": "OpenBMB's vision-language model with 35B total parameters and roughly 3B active per token. Understands images, documents, and multiple images, with native tool calling and optional thinking. The Q4_K_M download includes the 16x vision projector. Requires llama.cpp with MiniCPM-V 4.7 support (upstream PR #29416), which Gezel's bundled engine does not have yet; release builds hide it until one does.",
+  "tags": [
+    "openbmb",
+    "vision",
+    "multimodal",
+    "tools",
+    "moe"
+  ],
+  "category": "general",
+  "maintainer": {
+    "name": "OpenBMB",
+    "url": "https://huggingface.co/openbmb"
+  },
+  "version": "1.0.0",
+  "updatedAt": "2026-10-13T00:00:00Z",
+  "license": "Apache-2.0",
+  "licenseClass": "open",
+  "licenseShortName": "Apache 2.0",
+  "parameterSize": "35B",
+  "approxSizeBytes": 22285830848,
+  "supportsTools": true,
+  "contextWindow": 262144,
+  "upstream": "https://huggingface.co/openbmb/MiniCPM-V-4.7-35B-A3B",
+  "style": {
+    "family": "qwen",
+    "reasoningFormat": "think",
+    "toolCallFormat": "function-call"
+  },
+  "behaviors": [
+    "provider.merge-system-messages",
+    "reasoning.strip-think-tags",
+    "mcp.compact-tool-schemas",
+    "prompt.tool-cookbook-condensed",
+    "fabrication.detect-past-tense-no-tools",
+    "turn.preamble-folding",
+    "turn.ramble-detection"
+  ],
+  "tuning": {
+    "sampling": {
+      "temperature": 0.7,
+      "topP": 0.8,
+      "topK": 100,
+      "minP": 0,
+      "repetitionPenalty": 1.05,
+      "maxTokens": 8192
+    },
+    "reasoning": {
+      "enableThinking": false,
+      "thinkingBudget": 4096
+    },
+    "profiles": {
+      "thinking-general": {
+        "reasoning": {
+          "thinkingBudget": 4096,
+          "enableThinking": true
+        }
+      },
+      "thinking-coding": {
+        "sampling": {
+          "maxTokens": 16384
+        },
+        "reasoning": {
+          "thinkingBudget": 4096,
+          "enableThinking": true
+        }
+      },
+      "thinking-precise": {
+        "sampling": {
+          "seed": 0,
+          "temperature": 0.2
+        },
+        "reasoning": {
+          "thinkingBudget": 4096,
+          "enableThinking": true
+        }
+      },
+      "instruct": {
+        "reasoning": {
+          "enableThinking": false
+        }
+      },
+      "creative": {
+        "sampling": {
+          "topP": 0.9,
+          "temperature": 0.8
+        },
+        "reasoning": {
+          "enableThinking": false
+        }
+      }
+    }
+  },
+  "llamaCpp": {
+    "huggingfaceRepo": "openbmb/MiniCPM-V-4.7-35B-A3B-gguf",
+    "revision": "1e04eae556a075a1e78808a50266ed0b11de9bf1",
+    "filename": "MiniCPM-V-4_7-35B-A3B-Q4_K_M.gguf",
+    "sha256": "bf3d1c67f7843193e4483aebd26c5dea6b6d1f6a06ffa03c2387a6f3cc6d7df5",
+    "approxSizeBytes": 21166255008,
+    "quantization": "Q4_K_M",
+    "mmproj": {
+      "filename": "mmproj-model-f16.gguf",
+      "sha256": "bff3b688e14c300eca273092c37c50a0ddfcc2054e079af1a0480edeed5d20f1",
+      "sizeBytes": 1119575840
+    }
+  },
+  "mlx": {
+    "huggingfaceRepo": "mlx-community/MiniCPM-V-4.7-35B-A3B-4bit",
+    "revision": "0e3b706c1599ad95294122c67c77ce662de18794",
+    "quantization": "4bit",
+    "approxSizeBytes": 23548600508,
+    "files": [
+      {
+        "name": "chat_template.jinja",
+        "sha256": "c4531cacbf2a0ab65170433d7a4f371d11b7599cf96c04b192e6372432636c70",
+        "sizeBytes": 7250
+      },
+      {
+        "name": "config.json",
+        "sha256": "61335970616d168207bbf4d02daf87da7fe051ade19540d8711a8b37c1a06921",
+        "sizeBytes": 23933
+      },
+      {
+        "name": "generation_config.json",
+        "sha256": "7fa47217744f0e0918f6cf846e7541e4f322e6bc66917044cf2ad580699216c8",
+        "sizeBytes": 186
+      },
+      {
+        "name": "model-00001-of-00005.safetensors",
+        "sha256": "84a6955049b3d7428220c7f82dfd18701a8b932e99c8f2fee78a5c496d954636",
+        "sizeBytes": 5306512011
+      },
+      {
+        "name": "model-00002-of-00005.safetensors",
+        "sha256": "06ea042631cac78852e4283b12857c7b58f28dfa824d1cd6ba0bd9374701b5a4",
+        "sizeBytes": 5292106933
+      },
+      {
+        "name": "model-00003-of-00005.safetensors",
+        "sha256": "af2cca3162bc19f8de16cd3c76e169cb875a55b39786531d81fb55bc648035fe",
+        "sizeBytes": 5364936464
+      },
+      {
+        "name": "model-00004-of-00005.safetensors",
+        "sha256": "1c90bac77652fd7720556e73cdc66658b3daa4be0f205d4b8d2fabe3580c89c0",
+        "sizeBytes": 5352595712
+      },
+      {
+        "name": "model-00005-of-00005.safetensors",
+        "sha256": "4e774554cb00e35a0c1e043bae857e2e6f99efebb657a8279ec96aba1d441b52",
+        "sizeBytes": 2212214998
+      },
+      {
+        "name": "model.safetensors.index.json",
+        "sha256": "a369a2f73e06421915f7f1f04633f1db64c7fda170e5f5f3d1bec865cd1a8652",
+        "sizeBytes": 199017
+      },
+      {
+        "name": "preprocessor_config.json",
+        "sha256": "c64289d04f36e48d9d059e3a0069f54062b2ef6044a807adb448b98449cb54de",
+        "sizeBytes": 148
+      },
+      {
+        "name": "processor_config.json",
+        "sha256": "8540e97ca17871b46aca8639b05f26db4eb4ac39d271b705c9cdbc3e32abbe07",
+        "sizeBytes": 764
+      },
+      {
+        "name": "tokenizer.json",
+        "sha256": "e4608c5d7b3f4f30ba83be16a82cdac856b2e7191d40f0a08cbacc76c2a926ee",
+        "sizeBytes": 20001935
+      },
+      {
+        "name": "tokenizer_config.json",
+        "sha256": "be74f538ac586f2daac1ede56d421e915d196fba24171bc8fcd05256e42108bc",
+        "sizeBytes": 1157
+      }
+    ],
+    "disabledReason": "Requires MLX-VLM with MiniCPM-V 4.7 support (upstream PR #2499, conversion tested at e928101a144a3f11a27c7ef8d71b6d889b2d250c). Gezel pins mlx-vlm 0.7.1, which does not implement this architecture. Pending a compatible runtime and Apple Silicon validation."
+  }
+}

package/dist/catalog/chat-models/mi/minicpm-v4.7-35b-a3b-q4/versions/1.0.0/manifest.json

@@ -0,0 +1,94 @@
+{
+  "schemaVersion": 1,
+  "version": "1.0.0",
+  "releasedAt": "2026-10-13T00:00:00Z",
+  "minGezelVersion": "1.99999",
+  "approxSizeBytes": 22285830848,
+  "llamaCpp": {
+    "huggingfaceRepo": "openbmb/MiniCPM-V-4.7-35B-A3B-gguf",
+    "revision": "1e04eae556a075a1e78808a50266ed0b11de9bf1",
+    "filename": "MiniCPM-V-4_7-35B-A3B-Q4_K_M.gguf",
+    "sha256": "bf3d1c67f7843193e4483aebd26c5dea6b6d1f6a06ffa03c2387a6f3cc6d7df5",
+    "approxSizeBytes": 21166255008,
+    "quantization": "Q4_K_M",
+    "mmproj": {
+      "filename": "mmproj-model-f16.gguf",
+      "sha256": "bff3b688e14c300eca273092c37c50a0ddfcc2054e079af1a0480edeed5d20f1",
+      "sizeBytes": 1119575840
+    }
+  },
+  "mlx": {
+    "huggingfaceRepo": "mlx-community/MiniCPM-V-4.7-35B-A3B-4bit",
+    "revision": "0e3b706c1599ad95294122c67c77ce662de18794",
+    "quantization": "4bit",
+    "approxSizeBytes": 23548600508,
+    "files": [
+      {
+        "name": "chat_template.jinja",
+        "sha256": "c4531cacbf2a0ab65170433d7a4f371d11b7599cf96c04b192e6372432636c70",
+        "sizeBytes": 7250
+      },
+      {
+        "name": "config.json",
+        "sha256": "61335970616d168207bbf4d02daf87da7fe051ade19540d8711a8b37c1a06921",
+        "sizeBytes": 23933
+      },
+      {
+        "name": "generation_config.json",
+        "sha256": "7fa47217744f0e0918f6cf846e7541e4f322e6bc66917044cf2ad580699216c8",
+        "sizeBytes": 186
+      },
+      {
+        "name": "model-00001-of-00005.safetensors",
+        "sha256": "84a6955049b3d7428220c7f82dfd18701a8b932e99c8f2fee78a5c496d954636",
+        "sizeBytes": 5306512011
+      },
+      {
+        "name": "model-00002-of-00005.safetensors",
+        "sha256": "06ea042631cac78852e4283b12857c7b58f28dfa824d1cd6ba0bd9374701b5a4",
+        "sizeBytes": 5292106933
+      },
+      {
+        "name": "model-00003-of-00005.safetensors",
+        "sha256": "af2cca3162bc19f8de16cd3c76e169cb875a55b39786531d81fb55bc648035fe",
+        "sizeBytes": 5364936464
+      },
+      {
+        "name": "model-00004-of-00005.safetensors",
+        "sha256": "1c90bac77652fd7720556e73cdc66658b3daa4be0f205d4b8d2fabe3580c89c0",
+        "sizeBytes": 5352595712
+      },
+      {
+        "name": "model-00005-of-00005.safetensors",
+        "sha256": "4e774554cb00e35a0c1e043bae857e2e6f99efebb657a8279ec96aba1d441b52",
+        "sizeBytes": 2212214998
+      },
+      {
+        "name": "model.safetensors.index.json",
+        "sha256": "a369a2f73e06421915f7f1f04633f1db64c7fda170e5f5f3d1bec865cd1a8652",
+        "sizeBytes": 199017
+      },
+      {
+        "name": "preprocessor_config.json",
+        "sha256": "c64289d04f36e48d9d059e3a0069f54062b2ef6044a807adb448b98449cb54de",
+        "sizeBytes": 148
+      },
+      {
+        "name": "processor_config.json",
+        "sha256": "8540e97ca17871b46aca8639b05f26db4eb4ac39d271b705c9cdbc3e32abbe07",
+        "sizeBytes": 764
+      },
+      {
+        "name": "tokenizer.json",
+        "sha256": "e4608c5d7b3f4f30ba83be16a82cdac856b2e7191d40f0a08cbacc76c2a926ee",
+        "sizeBytes": 20001935
+      },
+      {
+        "name": "tokenizer_config.json",
+        "sha256": "be74f538ac586f2daac1ede56d421e915d196fba24171bc8fcd05256e42108bc",
+        "sizeBytes": 1157
+      }
+    ],
+    "disabledReason": "Requires MLX-VLM with MiniCPM-V 4.7 support (upstream PR #2499, conversion tested at e928101a144a3f11a27c7ef8d71b6d889b2d250c). Gezel pins mlx-vlm 0.7.1, which does not implement this architecture. Pending a compatible runtime and Apple Silicon validation."
+  }
+}

package/dist/catalog/chat-models/mi/minicpm5-1b-q4/manifest.json

@@ -20,6 +20,7 @@
   "version": "1.0.0",
   "updatedAt": "2026-09-27T00:00:00Z",
   "license": "Apache-2.0",
+  "licenseClass": "open",
   "parameterSize": "1B",
   "approxSizeBytes": 688065920,
   "supportsTools": true,

package/dist/catalog/chat-models/mi/minicpm5-2b-q4/manifest.json

@@ -20,6 +20,7 @@
   "version": "1.0.0",
   "updatedAt": "2026-09-27T00:00:00Z",
   "license": "Apache-2.0",
+  "licenseClass": "open",
   "parameterSize": "2B",
   "approxSizeBytes": 1561318368,
   "supportsTools": true,

package/dist/catalog/chat-models/qw/qwen3.5-0.8b-q4/manifest.json

@@ -19,6 +19,7 @@
   "version": "1.0.0",
   "updatedAt": "2026-09-27T00:00:00Z",
   "license": "Apache-2.0",
+  "licenseClass": "open",
   "parameterSize": "0.8B",
   "approxSizeBytes": 549698976,
   "supportsTools": true,