@bendyline/gilde 0.1.17 → 0.1.19

package/authoring/chat-models/gemma4-31b-q4.json

@@ -0,0 +1,81 @@
+{
+  "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 — 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": "1.2.1",
+  "updatedAt": "2026-07-28T00:00:00Z",
+  "license": "Apache-2.0",
+  "recoScore": 15,
+  "parameterSize": "30.7B",
+  "approxSizeBytes": 17287670048,
+  "supportsTools": true,
+  "contextWindow": 256000,
+  "upstream": "https://huggingface.co/google/gemma-4-31B-it",
+  "style": {
+    "family": "gemma",
+    "reasoningFormat": "channel",
+    "toolCallFormat": "function-call"
+  },
+  "behaviors": [
+    "reasoning.strip-channel-tags",
+    "reasoning.capture-pre-tool-prose",
+    "prompt.tool-cookbook-full",
+    "fabrication.detect-past-tense-no-tools",
+    "fabrication.detect-claim-without-tool",
+    "turn.preamble-folding",
+    "turn.ramble-detection",
+    "turn.auto-acknowledge-tool-errors",
+    "mcp.validate-ids-strict",
+    "parse.gemma-special-token"
+  ],
+  "tuning": {
+    "sampling": {
+      "temperature": 1,
+      "topP": 0.95,
+      "topK": 64,
+      "repetitionPenalty": 1.1,
+      "repetitionContext": 20
+    }
+  },
+  "ollama": {
+    "tag": "gemma4:31b-it-qat"
+  },
+  "llamaCpp": {
+    "huggingfaceRepo": "unsloth/gemma-4-31B-it-qat-GGUF",
+    "filename": "gemma-4-31B-it-qat-UD-Q4_K_XL.gguf",
+    "quantization": "UD-Q4_K_XL",
+    "residentBytes": 21081151219,
+    "draftModelFilename": "MTP/mtp-gemma-4-31B-it-Q4_0.gguf"
+  },
+  "releasePatch": {
+    "tuning": {
+      "engine": {
+        "llamaCpp": {
+          "spec": {
+            "mtp": true,
+            "nMax": 4
+          }
+        }
+      }
+    }
+  },
+  "mlx": {
+    "huggingfaceRepo": "mlx-community/gemma-4-31B-it-qat-4bit",
+    "quantization": "4bit"
+  }
+}

package/authoring/chat-models/gemma4-e2b-q4.json

@@ -0,0 +1,76 @@
+{
+  "id": "gemma4-e2b-q4",
+  "name": "Gemma 4 (E2B, Q4)",
+  "description": "Google's edge-optimized Gemma 4 at ~2B effective parameters. Multimodal (text + image + audio), native tool calling. Quantization-aware-trained 4-bit weights — the safe small-model default for any modern laptop.",
+  "tags": [
+    "google",
+    "multimodal",
+    "vision",
+    "audio",
+    "tools",
+    "small",
+    "qat"
+  ],
+  "category": "general",
+  "maintainer": {
+    "name": "Google",
+    "url": "https://huggingface.co/google/gemma-4-E2B-it"
+  },
+  "version": "1.2.0",
+  "updatedAt": "2026-07-31T00:00:00Z",
+  "license": "Apache-2.0",
+  "recoScore": 15,
+  "parameterSize": "2.3B",
+  "approxSizeBytes": 2620370976,
+  "supportsTools": true,
+  "contextWindow": 128000,
+  "upstream": "https://huggingface.co/google/gemma-4-E2B-it",
+  "style": {
+    "family": "gemma",
+    "reasoningFormat": "channel",
+    "toolCallFormat": "function-call"
+  },
+  "behaviors": [
+    "reasoning.strip-channel-tags",
+    "prompt.tool-cookbook-full",
+    "fabrication.detect-past-tense-no-tools",
+    "fabrication.detect-claim-without-tool",
+    "mcp.relax-required-fields",
+    "mcp.default-missing-fields",
+    {
+      "id": "turn.continuation-budget",
+      "config": {
+        "count": 4
+      }
+    },
+    "turn.preamble-folding",
+    "turn.ramble-detection",
+    "parse.gemma-special-token"
+  ],
+  "ollama": {
+    "tag": "gemma4:e2b"
+  },
+  "llamaCpp": {
+    "huggingfaceRepo": "unsloth/gemma-4-E2B-it-qat-GGUF",
+    "filename": "gemma-4-E2B-it-qat-UD-Q4_K_XL.gguf",
+    "quantization": "UD-Q4_K_XL",
+    "residentBytes": 3731251642,
+    "draftModelFilename": "MTP/mtp-gemma-4-E2B-it-Q4_0.gguf"
+  },
+  "releasePatch": {
+    "tuning": {
+      "engine": {
+        "llamaCpp": {
+          "spec": {
+            "mtp": true,
+            "nMax": 4
+          }
+        }
+      }
+    }
+  },
+  "mlx": {
+    "huggingfaceRepo": "mlx-community/gemma-4-E2B-it-qat-4bit",
+    "quantization": "4bit"
+  }
+}

package/authoring/chat-models/gemma4-e4b-q4.json

@@ -0,0 +1,71 @@
+{
+  "id": "gemma4-e4b-q4",
+  "name": "Gemma 4 (E4B, Q4)",
+  "description": "Google's edge-optimized Gemma 4 at ~4B effective parameters. Multimodal (text + image + audio), native tool calling. Quantization-aware-trained 4-bit weights — a strong compact default that leaves room to work on a 16GB laptop.",
+  "tags": [
+    "google",
+    "multimodal",
+    "vision",
+    "audio",
+    "tools",
+    "qat"
+  ],
+  "category": "general",
+  "maintainer": {
+    "name": "Google",
+    "url": "https://huggingface.co/google/gemma-4-E4B-it"
+  },
+  "version": "1.2.0",
+  "updatedAt": "2026-07-31T00:00:00Z",
+  "license": "Apache-2.0",
+  "recoScore": 15,
+  "parameterSize": "4.5B",
+  "approxSizeBytes": 4215695776,
+  "supportsTools": true,
+  "contextWindow": 128000,
+  "upstream": "https://huggingface.co/google/gemma-4-E4B-it",
+  "style": {
+    "family": "gemma",
+    "reasoningFormat": "channel",
+    "toolCallFormat": "function-call"
+  },
+  "behaviors": [
+    "reasoning.strip-channel-tags",
+    "prompt.tool-cookbook-condensed",
+    "fabrication.detect-past-tense-no-tools",
+    "fabrication.detect-claim-without-tool",
+    "mcp.relax-required-fields",
+    "mcp.default-missing-fields",
+    "turn.preamble-folding",
+    "turn.ramble-detection",
+    "parse.gemma-special-token"
+  ],
+  "ollama": {
+    "tag": "gemma4:e4b"
+  },
+  "llamaCpp": {
+    "huggingfaceRepo": "unsloth/gemma-4-E4B-it-qat-GGUF",
+    "filename": "gemma-4-E4B-it-qat-UD-Q4_K_XL.gguf",
+    "quantization": "UD-Q4_K_XL",
+    "residentBytes": 5940328250,
+    "draftModelFilename": "MTP/mtp-gemma-4-E4B-it-Q4_0.gguf"
+  },
+  "releasePatch": {
+    "tuning": {
+      "engine": {
+        "llamaCpp": {
+          "spec": {
+            "type": null,
+            "draftModelId": null,
+            "mtp": true,
+            "nMax": 4
+          }
+        }
+      }
+    }
+  },
+  "mlx": {
+    "huggingfaceRepo": "mlx-community/gemma-4-E4B-it-qat-4bit",
+    "quantization": "4bit"
+  }
+}

package/authoring/chat-models/glm-5.2-754b-q2.json

@@ -0,0 +1,60 @@
+{
+  "id": "glm-5.2-754b-q2",
+  "name": "GLM 5.2 (IQ2_XXS)",
+  "description": "antirez's routed 2-bit community GGUF of Z.ai's GLM 5.2 — a 754B-parameter Mixture-of-Experts (256 experts, 8 active per token, ~40B active) with a 1M-token context and DeepSeek-style sparse attention. The routed experts are crushed to IQ2_XXS (block 78 in Q2_K) while attention, the shared expert, embeddings, and the output head stay at Q8-class precision, landing the file at ~197 GB. Only ~20 GB of that is non-routed weight that must stay resident: DwarfStar streams the remaining ~177 GB of experts from SSD, so it runs on a 128 GB machine — but it is a very large download and a slow first turn. Built for antirez's `ds4` engine; these quants do not load on stock llama.cpp.",
+  "tags": [
+    "glm",
+    "reasoning",
+    "moe",
+    "tools",
+    "2-bit",
+    "iq2",
+    "long-context",
+    "large"
+  ],
+  "category": "reasoning",
+  "maintainer": {
+    "name": "antirez",
+    "url": "https://huggingface.co/antirez/glm-5.2-gguf"
+  },
+  "version": "1.0.0",
+  "updatedAt": "2026-07-25T00:00:00Z",
+  "license": "MIT",
+  "parameterSize": "754B",
+  "approxSizeBytes": 211075856448,
+  "supportsTools": true,
+  "contextWindow": 1048576,
+  "upstream": "https://huggingface.co/zai-org/GLM-5.2",
+  "style": {
+    "family": "glm",
+    "reasoningFormat": "think",
+    "toolCallFormat": "function-call"
+  },
+  "behaviors": [
+    "reasoning.strip-think-tags",
+    "prompt.tool-cookbook-condensed",
+    "mcp.compact-tool-schemas",
+    "fabrication.detect-past-tense-no-tools",
+    "turn.preamble-folding"
+  ],
+  "tuning": {
+    "sampling": {
+      "temperature": 1,
+      "topP": 0.95,
+      "maxTokens": 16384
+    },
+    "reasoning": {
+      "enableThinking": true,
+      "thinkingBudget": 4096
+    }
+  },
+  "ds4": {
+    "huggingfaceRepo": "antirez/glm-5.2-gguf",
+    "filename": "GLM-5.2-UD-IQ2_XXS_RoutedIQ2XXS_blk78Q2K.gguf",
+    "quantization": "IQ2_XXS",
+    "residentBytes": 61363217408,
+    "cacheExpertsBytes": 34359738368,
+    "ssdStreaming": true,
+    "maxLaunchCtx": 65536
+  }
+}

package/authoring/chat-models/gpt-oss-120b-q4.json

@@ -0,0 +1,48 @@
+{
+  "id": "gpt-oss-120b-q4",
+  "name": "GPT-OSS (120B, MoE)",
+  "description": "OpenAI's open-weight 117B-parameter Mixture-of-Experts model (~5.1B active per token). Configurable reasoning effort (low/medium/high), full chain-of-thought access, and native function-calling. Frontier-class on-device reasoning for workstations and Mac Studio-class hardware — needs ~80GB of usable memory.",
+  "tags": [
+    "openai",
+    "tools",
+    "reasoning",
+    "moe",
+    "open-weights",
+    "large"
+  ],
+  "category": "reasoning",
+  "maintainer": {
+    "name": "OpenAI",
+    "url": "https://ollama.com/library/gpt-oss"
+  },
+  "version": "1.1.0",
+  "updatedAt": "2026-05-17T00:00:00Z",
+  "license": "Apache-2.0",
+  "parameterSize": "117B",
+  "approxSizeBytes": 65000000000,
+  "supportsTools": true,
+  "contextWindow": 131072,
+  "upstream": "https://huggingface.co/openai/gpt-oss-120b",
+  "style": {
+    "family": "gpt-oss",
+    "reasoningFormat": "channel",
+    "toolCallFormat": "function-call"
+  },
+  "behaviors": [
+    "reasoning.strip-channel-tags",
+    "fabrication.detect-past-tense-no-tools",
+    "turn.preamble-folding"
+  ],
+  "ollama": {
+    "tag": "gpt-oss:120b"
+  },
+  "llamaCpp": {
+    "huggingfaceRepo": "bartowski/openai_gpt-oss-120b-GGUF",
+    "shardsDir": "openai_gpt-oss-120b-Q4_K_M",
+    "quantization": "Q4_K_M"
+  },
+  "mlx": {
+    "huggingfaceRepo": "mlx-community/gpt-oss-120b-MXFP4-Q4",
+    "quantization": "MXFP4-Q4"
+  }
+}

package/authoring/chat-models/gpt-oss-20b-q4.json

@@ -0,0 +1,47 @@
+{
+  "id": "gpt-oss-20b-q4",
+  "name": "GPT-OSS (20B, MoE)",
+  "description": "OpenAI's open-weight 20B-parameter Mixture-of-Experts model (~3.6B active per token). Configurable reasoning effort (low/medium/high), full chain-of-thought access, and native function-calling. Lower-latency frontier-class on-device option for machines with 16GB+ memory.",
+  "tags": [
+    "openai",
+    "tools",
+    "reasoning",
+    "moe",
+    "open-weights"
+  ],
+  "category": "reasoning",
+  "maintainer": {
+    "name": "OpenAI",
+    "url": "https://ollama.com/library/gpt-oss"
+  },
+  "version": "1.1.0",
+  "updatedAt": "2026-04-26T00:00:00Z",
+  "license": "Apache-2.0",
+  "parameterSize": "20B",
+  "approxSizeBytes": 12000000000,
+  "supportsTools": true,
+  "contextWindow": 131072,
+  "upstream": "https://ollama.com/library/gpt-oss",
+  "style": {
+    "family": "gpt-oss",
+    "reasoningFormat": "channel",
+    "toolCallFormat": "function-call"
+  },
+  "behaviors": [
+    "reasoning.strip-channel-tags",
+    "fabrication.detect-past-tense-no-tools",
+    "turn.preamble-folding"
+  ],
+  "ollama": {
+    "tag": "gpt-oss:20b"
+  },
+  "llamaCpp": {
+    "huggingfaceRepo": "bartowski/openai_gpt-oss-20b-GGUF",
+    "filename": "openai_gpt-oss-20b-Q4_K_M.gguf",
+    "quantization": "Q4_K_M"
+  },
+  "mlx": {
+    "huggingfaceRepo": "mlx-community/gpt-oss-20b-MXFP4-Q4",
+    "quantization": "MXFP4-Q4"
+  }
+}

package/authoring/chat-models/laguna-s-2.1-118b-q4.json

@@ -0,0 +1,127 @@
+{
+  "id": "laguna-s-2.1-118b-q4",
+  "name": "Laguna S 2.1 (118B-A8B, Q4)",
+  "description": "Poolside's Laguna S 2.1, a 118B-parameter Mixture-of-Experts model with about 8B active parameters per token, built for agentic coding and long-horizon software work. It supports native tool calling and interleaved reasoning between tool calls. Choose the official imatrix Q4_K_M GGUF (about 75GB) for llama.cpp, or the calibrated oQ4e MLX build (about 64GB) for native Apple Silicon inference on a 96GB-or-larger Mac. Gezel defaults to Poolside's quality-recommended 256K context window; the underlying checkpoint can be extended to 1M context with explicit YaRN overrides.",
+  "tags": [
+    "poolside",
+    "coding",
+    "agentic",
+    "tools",
+    "reasoning",
+    "long-context",
+    "moe",
+    "large",
+    "open-weights",
+    "apple-silicon"
+  ],
+  "category": "coding",
+  "maintainer": {
+    "name": "Poolside",
+    "url": "https://huggingface.co/poolside/Laguna-S-2.1"
+  },
+  "version": "1.0.1",
+  "updatedAt": "2026-07-28T00:00:00Z",
+  "license": "OpenMDW-1.1",
+  "licenseClass": "open",
+  "licenseShortName": "OpenMDW 1.1",
+  "licenseUrl": "https://huggingface.co/poolside/Laguna-S-2.1/blob/main/LICENSE.md",
+  "parameterSize": "118B",
+  "approxSizeBytes": 96031829760,
+  "supportsTools": true,
+  "contextWindow": 262144,
+  "upstream": "https://huggingface.co/poolside/Laguna-S-2.1",
+  "style": {
+    "family": "other",
+    "reasoningFormat": "think",
+    "toolCallFormat": "function-call"
+  },
+  "behaviors": [
+    "reasoning.strip-think-tags",
+    "fabrication.detect-past-tense-no-tools",
+    "turn.preamble-folding",
+    "turn.ramble-detection",
+    "tools.mlx-grammar"
+  ],
+  "tuning": {
+    "sampling": {
+      "temperature": 0.7,
+      "topP": 0.95,
+      "maxTokens": 8192
+    },
+    "samplingWhenThinking": {
+      "temperature": 0.7,
+      "topP": 0.95
+    },
+    "reasoning": {
+      "enableThinking": true,
+      "thinkingBudget": 4096
+    },
+    "profiles": {
+      "thinking-general": {
+        "sampling": {
+          "temperature": 0.7,
+          "topP": 0.95,
+          "maxTokens": 12288
+        },
+        "reasoning": {
+          "enableThinking": true,
+          "thinkingBudget": 4096
+        }
+      },
+      "thinking-coding": {
+        "sampling": {
+          "temperature": 0.6,
+          "topP": 0.95,
+          "maxTokens": 16384
+        },
+        "reasoning": {
+          "enableThinking": true,
+          "thinkingBudget": 6144
+        }
+      },
+      "thinking-precise": {
+        "sampling": {
+          "temperature": 0.4,
+          "topP": 0.9,
+          "seed": 0,
+          "maxTokens": 6144
+        },
+        "reasoning": {
+          "enableThinking": true,
+          "thinkingBudget": 4096
+        }
+      },
+      "instruct": {
+        "sampling": {
+          "temperature": 0.7,
+          "topP": 0.95,
+          "maxTokens": 4096
+        },
+        "reasoning": {
+          "enableThinking": false
+        }
+      },
+      "creative": {
+        "sampling": {
+          "temperature": 1,
+          "topP": 0.95,
+          "maxTokens": 8192
+        },
+        "reasoning": {
+          "enableThinking": false
+        }
+      }
+    }
+  },
+  "llamaCpp": {
+    "huggingfaceRepo": "poolside/Laguna-S-2.1-GGUF",
+    "filename": "laguna-s-2.1-Q4_K_M.gguf",
+    "quantization": "Q4_K_M",
+    "residentBytes": 115238195712
+  },
+  "mlx": {
+    "huggingfaceRepo": "mlx-community/Laguna-S-2.1-oQ4e",
+    "quantization": "oQ4e",
+    "residentBytes": 70000000000
+  }
+}

package/authoring/chat-models/laguna-s-2.1-118b-q6.json

@@ -0,0 +1,121 @@
+{
+  "id": "laguna-s-2.1-118b-q6",
+  "name": "Laguna S 2.1 (118B-A8B, Q6 MLX)",
+  "description": "Poolside's Laguna S 2.1 in a calibrated oQ6e MLX quantization for native Apple Silicon inference. The model has 118B total parameters with about 8B active per token and is built for agentic coding, long-horizon software work, native tool calling, and interleaved reasoning. This MLX-only build is about 92GB on disk, measured about 90GB peak at 64K context, and targets 128GB Macs. Gezel defaults to Poolside's quality-recommended 256K context window; the underlying checkpoint can be extended to 1M context with explicit YaRN overrides.",
+  "tags": [
+    "poolside",
+    "coding",
+    "agentic",
+    "tools",
+    "reasoning",
+    "long-context",
+    "moe",
+    "large",
+    "open-weights",
+    "apple-silicon"
+  ],
+  "category": "coding",
+  "maintainer": {
+    "name": "Poolside",
+    "url": "https://huggingface.co/poolside/Laguna-S-2.1"
+  },
+  "version": "1.0.0",
+  "updatedAt": "2026-07-22T00:00:00Z",
+  "license": "OpenMDW-1.1",
+  "licenseClass": "open",
+  "licenseShortName": "OpenMDW 1.1",
+  "licenseUrl": "https://huggingface.co/poolside/Laguna-S-2.1/blob/main/LICENSE.md",
+  "parameterSize": "118B",
+  "approxSizeBytes": 92000000000,
+  "supportsTools": true,
+  "contextWindow": 262144,
+  "upstream": "https://huggingface.co/poolside/Laguna-S-2.1",
+  "style": {
+    "family": "other",
+    "reasoningFormat": "think",
+    "toolCallFormat": "function-call"
+  },
+  "behaviors": [
+    "reasoning.strip-think-tags",
+    "fabrication.detect-past-tense-no-tools",
+    "turn.preamble-folding",
+    "turn.ramble-detection",
+    "tools.mlx-grammar"
+  ],
+  "tuning": {
+    "sampling": {
+      "temperature": 0.7,
+      "topP": 0.95,
+      "maxTokens": 8192
+    },
+    "samplingWhenThinking": {
+      "temperature": 0.7,
+      "topP": 0.95
+    },
+    "reasoning": {
+      "enableThinking": true,
+      "thinkingBudget": 4096
+    },
+    "profiles": {
+      "thinking-general": {
+        "sampling": {
+          "temperature": 0.7,
+          "topP": 0.95,
+          "maxTokens": 12288
+        },
+        "reasoning": {
+          "enableThinking": true,
+          "thinkingBudget": 4096
+        }
+      },
+      "thinking-coding": {
+        "sampling": {
+          "temperature": 0.6,
+          "topP": 0.95,
+          "maxTokens": 16384
+        },
+        "reasoning": {
+          "enableThinking": true,
+          "thinkingBudget": 6144
+        }
+      },
+      "thinking-precise": {
+        "sampling": {
+          "temperature": 0.4,
+          "topP": 0.9,
+          "seed": 0,
+          "maxTokens": 6144
+        },
+        "reasoning": {
+          "enableThinking": true,
+          "thinkingBudget": 4096
+        }
+      },
+      "instruct": {
+        "sampling": {
+          "temperature": 0.7,
+          "topP": 0.95,
+          "maxTokens": 4096
+        },
+        "reasoning": {
+          "enableThinking": false
+        }
+      },
+      "creative": {
+        "sampling": {
+          "temperature": 1,
+          "topP": 0.95,
+          "maxTokens": 8192
+        },
+        "reasoning": {
+          "enableThinking": false
+        }
+      }
+    }
+  },
+  "mlx": {
+    "huggingfaceRepo": "mlx-community/Laguna-S-2.1-oQ6e",
+    "quantization": "oQ6e",
+    "residentBytes": 99000000000
+  }
+}