@bendyline/gilde 0.1.18 → 0.1.20

package/authoring/chat-models/deepseek-v4-flash-284b-q2q4.json

@@ -0,0 +1,111 @@
+{
+  "id": "deepseek-v4-flash-284b-q2q4",
+  "name": "DeepSeek V4 Flash (mixed 2/4-bit)",
+  "description": "antirez's mixed-precision community GGUF of DeepSeek V4 Flash 0731 — the July 2026 re-post-trained 284B-parameter Mixture-of-Experts (~13B active per token) with 1M-token context and Compressed Sparse Attention. Layers 37-42 keep Q4_K routed experts while the remaining expert layers drop to IQ2_XXS gate/up + Q2_K down, landing at ~98 GB. This is the quality/size sweet spot antirez recommends for 128 GB MacBooks: noticeably better than the straight 2-bit build without the ~156 GB footprint of the full Q4_K one. Produced for antirez's `ds4` inference engine; whether it loads on stock llama depends on its DeepSeek-V4 support.",
+  "tags": [
+    "deepseek",
+    "reasoning",
+    "agentic",
+    "moe",
+    "tools",
+    "long-context",
+    "large",
+    "mixed-precision",
+    "iq2",
+    "q4"
+  ],
+  "category": "reasoning",
+  "maintainer": {
+    "name": "antirez",
+    "url": "https://huggingface.co/antirez/deepseek-v4-gguf"
+  },
+  "version": "1.0.0",
+  "updatedAt": "2026-08-02T00:00:00Z",
+  "license": "MIT",
+  "licenseClass": "open",
+  "licenseShortName": "MIT",
+  "parameterSize": "284B",
+  "approxSizeBytes": 97591747456,
+  "supportsTools": true,
+  "contextWindow": 1000000,
+  "upstream": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash",
+  "style": {
+    "family": "deepseek",
+    "reasoningFormat": "think",
+    "toolCallFormat": "function-call"
+  },
+  "behaviors": [
+    "reasoning.strip-think-tags",
+    "prompt.tool-cookbook-condensed",
+    "prompt.meester-build-prelude",
+    "mcp.compact-tool-schemas",
+    "fabrication.detect-past-tense-no-tools",
+    "turn.preamble-folding"
+  ],
+  "tuning": {
+    "sampling": {
+      "maxTokens": 16384
+    },
+    "reasoning": {
+      "enableThinking": true,
+      "thinkingBudget": 4096
+    },
+    "profiles": {
+      "thinking-general": {
+        "sampling": {
+          "temperature": 0.6,
+          "topP": 0.95,
+          "maxTokens": 16384
+        }
+      },
+      "thinking-coding": {
+        "sampling": {
+          "temperature": 0.3,
+          "topP": 0.95,
+          "maxTokens": 16384
+        }
+      },
+      "thinking-precise": {
+        "sampling": {
+          "temperature": 0.1,
+          "topP": 0.9,
+          "maxTokens": 16384,
+          "seed": 0
+        }
+      },
+      "instruct": {
+        "sampling": {
+          "temperature": 0.7,
+          "topP": 0.95,
+          "maxTokens": 16384
+        },
+        "reasoning": {
+          "enableThinking": false
+        }
+      },
+      "creative": {
+        "sampling": {
+          "temperature": 1,
+          "topP": 1,
+          "maxTokens": 16384
+        },
+        "reasoning": {
+          "enableThinking": false
+        }
+      }
+    }
+  },
+  "llamaCpp": {
+    "huggingfaceRepo": "antirez/deepseek-v4-gguf",
+    "filename": "DeepSeek-V4-Flash-Layers37-42Q4KExperts-OtherExpertLayersIQ2XXSGateUp-Q2KDown-AProjQ8-SExpQ8-OutQ8-chat-v2-imatrix-fixed-0731.gguf",
+    "quantization": "IQ2_XXS/Q4_K"
+  },
+  "ds4": {
+    "huggingfaceRepo": "antirez/deepseek-v4-gguf",
+    "filename": "DeepSeek-V4-Flash-Layers37-42Q4KExperts-OtherExpertLayersIQ2XXSGateUp-Q2KDown-AProjQ8-SExpQ8-OutQ8-chat-v2-imatrix-fixed-0731.gguf",
+    "quantization": "IQ2_XXS/Q4_K",
+    "residentBytes": 45097156608,
+    "cacheExpertsBytes": 38654705664,
+    "ssdStreaming": true
+  }
+}

package/authoring/chat-models/deepseek-v4-flash-284b-q4.json

@@ -0,0 +1,113 @@
+{
+  "id": "deepseek-v4-flash-284b-q4",
+  "name": "DeepSeek V4 Flash (FP4)",
+  "description": "DeepSeek V4 Flash — 284B-parameter Mixture-of-Experts (~13B active per token) with 1M-token context, Compressed Sparse Attention, and configurable Non-think / Think-High / Think-Max reasoning modes. On the `ds4` engine this is the July 2026 0731 re-post-trained checkpoint, which is substantially stronger at agentic and coding work than the original release. The llama.cpp (native FP4+FP8) and MLX sources still carry the original V4 Flash weights — upstream has not republished those builds for 0731 — so pick the ds4 source to get the newer checkpoint. Frontier-tier on-device reasoning for workstations and Mac Studio-class hardware — needs ~200 GB of usable memory.",
+  "tags": [
+    "deepseek",
+    "reasoning",
+    "agentic",
+    "moe",
+    "tools",
+    "fp4",
+    "long-context",
+    "large"
+  ],
+  "category": "reasoning",
+  "maintainer": {
+    "name": "DeepSeek",
+    "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash"
+  },
+  "version": "1.1.0",
+  "updatedAt": "2026-08-02T00:00:00Z",
+  "license": "MIT",
+  "licenseClass": "open",
+  "licenseShortName": "MIT",
+  "parameterSize": "284B",
+  "approxSizeBytes": 156148189760,
+  "supportsTools": true,
+  "contextWindow": 1000000,
+  "upstream": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash",
+  "style": {
+    "family": "deepseek",
+    "reasoningFormat": "think",
+    "toolCallFormat": "function-call"
+  },
+  "behaviors": [
+    "reasoning.strip-think-tags",
+    "prompt.tool-cookbook-condensed",
+    "prompt.meester-build-prelude",
+    "mcp.compact-tool-schemas",
+    "fabrication.detect-past-tense-no-tools",
+    "turn.preamble-folding"
+  ],
+  "tuning": {
+    "sampling": {
+      "maxTokens": 16384
+    },
+    "reasoning": {
+      "enableThinking": true,
+      "thinkingBudget": 4096
+    },
+    "profiles": {
+      "thinking-general": {
+        "sampling": {
+          "temperature": 0.6,
+          "topP": 0.95,
+          "maxTokens": 16384
+        }
+      },
+      "thinking-coding": {
+        "sampling": {
+          "temperature": 0.3,
+          "topP": 0.95,
+          "maxTokens": 16384
+        }
+      },
+      "thinking-precise": {
+        "sampling": {
+          "temperature": 0.1,
+          "topP": 0.9,
+          "maxTokens": 16384,
+          "seed": 0
+        }
+      },
+      "instruct": {
+        "sampling": {
+          "temperature": 0.7,
+          "topP": 0.95,
+          "maxTokens": 16384
+        },
+        "reasoning": {
+          "enableThinking": false
+        }
+      },
+      "creative": {
+        "sampling": {
+          "temperature": 1,
+          "topP": 1,
+          "maxTokens": 16384
+        },
+        "reasoning": {
+          "enableThinking": false
+        }
+      }
+    }
+  },
+  "llamaCpp": {
+    "huggingfaceRepo": "nsparks/DeepSeek-V4-Flash-FP4-FP8-GGUF",
+    "filename": "DeepSeek-V4-Flash-FP4-FP8-native.gguf",
+    "quantization": "FP4-FP8"
+  },
+  "mlx": {
+    "huggingfaceRepo": "mlx-community/deepseek-ai-DeepSeek-V4-Flash-4bit",
+    "quantization": "4bit"
+  },
+  "ds4": {
+    "huggingfaceRepo": "antirez/deepseek-v4-gguf",
+    "filename": "DeepSeek-V4-Flash-Q4KExperts-F16HC-F16Compressor-F16Indexer-Q8Attn-Q8Shared-Q8Out-chat-v2-imatrix-0731.gguf",
+    "quantization": "Q4_K",
+    "residentBytes": 85899345920,
+    "cacheExpertsBytes": 68719476736,
+    "ssdStreaming": true
+  }
+}

package/authoring/chat-models/gemma4-12b-q4.json

@@ -0,0 +1,190 @@
+{
+  "id": "gemma4-12b-q4",
+  "name": "Gemma 4 (12B, Q4)",
+  "description": "Google's Gemma 4 at 12B parameters — an encoder-free unified multimodal model (text + image + audio + video) with native tool calling, configurable thinking, and a 256K context window. Apache-2.0 licensed.",
+  "tags": [
+    "google",
+    "multimodal",
+    "vision",
+    "audio",
+    "tools",
+    "agentic",
+    "qat"
+  ],
+  "category": "general",
+  "maintainer": {
+    "name": "Google",
+    "url": "https://huggingface.co/google/gemma-4-12B-it"
+  },
+  "version": "1.1.2",
+  "updatedAt": "2026-07-28T00:00:00Z",
+  "license": "Apache-2.0",
+  "recoScore": 15,
+  "parameterSize": "12B",
+  "approxSizeBytes": 6716356800,
+  "supportsTools": true,
+  "contextWindow": 256000,
+  "upstream": "https://huggingface.co/google/gemma-4-12B-it",
+  "style": {
+    "family": "gemma",
+    "reasoningFormat": "channel",
+    "toolCallFormat": "function-call"
+  },
+  "behaviors": [
+    "reasoning.strip-channel-tags",
+    "reasoning.capture-pre-tool-prose",
+    "prompt.tool-cookbook-full",
+    "prompt.meester-build-prelude",
+    "fabrication.detect-past-tense-no-tools",
+    "fabrication.detect-claim-without-tool",
+    "turn.preamble-folding",
+    {
+      "id": "turn.ramble-detection",
+      "config": {
+        "coldThreshold": 6000,
+        "postActionThreshold": 1200
+      }
+    },
+    "turn.auto-acknowledge-tool-errors",
+    "turn.permission-stall",
+    {
+      "id": "turn.continuation-budget",
+      "config": {
+        "count": 4
+      }
+    },
+    {
+      "id": "turn.single-tool-per-turn",
+      "config": {
+        "meesterOnly": true
+      }
+    },
+    "mcp.compact-tool-schemas",
+    "mcp.relax-required-fields",
+    "mcp.default-missing-fields",
+    "mcp.validate-ids-strict",
+    "parse.gemma-special-token",
+    "prompt.prefer-writefile-edits",
+    "prompt.terse-visible-reply"
+  ],
+  "tuning": {
+    "sampling": {
+      "temperature": 1,
+      "topP": 0.95,
+      "topK": 64,
+      "repetitionPenalty": 1.1,
+      "repetitionContext": 64,
+      "dry": {
+        "multiplier": 0.8,
+        "base": 1.75,
+        "allowedLength": 2
+      },
+      "maxTokens": 8192
+    },
+    "reasoning": {
+      "enableThinking": true,
+      "thinkingBudget": 256
+    },
+    "profiles": {
+      "thinking-general": {
+        "sampling": {
+          "temperature": 0.8,
+          "topP": 0.95,
+          "topK": 64,
+          "repetitionPenalty": 1.1,
+          "repetitionContext": 64,
+          "dry": {
+            "multiplier": 0.8,
+            "base": 1.75,
+            "allowedLength": 2
+          },
+          "maxTokens": 2048
+        }
+      },
+      "thinking-coding": {
+        "sampling": {
+          "temperature": 0.6,
+          "topP": 0.9,
+          "topK": 40,
+          "repetitionPenalty": 1.05,
+          "repetitionContext": 64,
+          "dry": {
+            "multiplier": 0.8,
+            "base": 1.75,
+            "allowedLength": 2
+          },
+          "maxTokens": 2048
+        }
+      },
+      "thinking-precise": {
+        "sampling": {
+          "temperature": 0.3,
+          "topP": 0.9,
+          "topK": 20,
+          "repetitionPenalty": 1.05,
+          "repetitionContext": 64,
+          "dry": {
+            "multiplier": 0.8,
+            "base": 1.75,
+            "allowedLength": 2
+          },
+          "maxTokens": 4096,
+          "seed": 0
+        }
+      },
+      "instruct": {
+        "sampling": {
+          "temperature": 0.7,
+          "topP": 0.95,
+          "topK": 64,
+          "repetitionPenalty": 1.1,
+          "repetitionContext": 64,
+          "dry": {
+            "multiplier": 0.8,
+            "base": 1.75,
+            "allowedLength": 2
+          },
+          "maxTokens": 4096
+        }
+      },
+      "creative": {
+        "sampling": {
+          "temperature": 1.1,
+          "topP": 0.95,
+          "topK": 64,
+          "repetitionPenalty": 1,
+          "repetitionContext": 20,
+          "maxTokens": 8192
+        }
+      }
+    }
+  },
+  "evalHints": {
+    "sniffThresholds": {
+      "inlineJsMinBytes": 1000
+    }
+  },
+  "llamaCpp": {
+    "huggingfaceRepo": "unsloth/gemma-4-12B-it-qat-GGUF",
+    "filename": "gemma-4-12B-it-qat-UD-Q4_K_XL.gguf",
+    "quantization": "UD-Q4_K_XL",
+    "residentBytes": 8364078720,
+    "draftModelFilename": "MTP/mtp-gemma-4-12B-it-Q4_0.gguf"
+  },
+  "releasePatch": {
+    "tuning": {
+      "engine": {
+        "llamaCpp": {
+          "spec": {
+            "mtp": true,
+            "nMax": 4
+          }
+        }
+      }
+    }
+  },
+  "mlx": {
+    "huggingfaceRepo": "mlx-community/gemma-4-12B-it-qat-4bit",
+    "quantization": "4bit"
+  }
+}

package/authoring/chat-models/gemma4-12b-q8.json

@@ -0,0 +1,164 @@
+{
+  "id": "gemma4-12b-q8",
+  "name": "Gemma 4 (12B, Q8)",
+  "description": "Google's Gemma 4 at 12B parameters, 8-bit quantization for higher fidelity than the Q4 build (at roughly double the memory footprint). Encoder-free unified multimodal model (text + image + audio + video), native tool calling, configurable thinking, 256K context. Apache-2.0 licensed.",
+  "tags": [
+    "google",
+    "multimodal",
+    "vision",
+    "audio",
+    "tools",
+    "agentic"
+  ],
+  "category": "general",
+  "maintainer": {
+    "name": "Google",
+    "url": "https://huggingface.co/google/gemma-4-12B-it"
+  },
+  "version": "1.0.2",
+  "updatedAt": "2026-07-28T00:00:00Z",
+  "license": "Apache-2.0",
+  "recoScore": 15,
+  "parameterSize": "12B",
+  "approxSizeBytes": 12669646976,
+  "supportsTools": true,
+  "contextWindow": 256000,
+  "upstream": "https://huggingface.co/google/gemma-4-12B-it",
+  "style": {
+    "family": "gemma",
+    "reasoningFormat": "channel",
+    "toolCallFormat": "function-call"
+  },
+  "behaviors": [
+    "reasoning.strip-channel-tags",
+    "reasoning.capture-pre-tool-prose",
+    "prompt.tool-cookbook-full",
+    "prompt.meester-build-prelude",
+    "fabrication.detect-past-tense-no-tools",
+    "fabrication.detect-claim-without-tool",
+    "turn.preamble-folding",
+    {
+      "id": "turn.ramble-detection",
+      "config": {
+        "coldThreshold": 6000,
+        "postActionThreshold": 1200
+      }
+    },
+    "turn.auto-acknowledge-tool-errors",
+    "turn.permission-stall",
+    {
+      "id": "turn.continuation-budget",
+      "config": {
+        "count": 4
+      }
+    },
+    {
+      "id": "turn.single-tool-per-turn",
+      "config": {
+        "meesterOnly": true
+      }
+    },
+    "mcp.compact-tool-schemas",
+    "mcp.relax-required-fields",
+    "mcp.default-missing-fields",
+    "mcp.validate-ids-strict",
+    "parse.gemma-special-token",
+    "prompt.prefer-writefile-edits",
+    "prompt.terse-visible-reply"
+  ],
+  "tuning": {
+    "sampling": {
+      "temperature": 1,
+      "topP": 0.95,
+      "topK": 64,
+      "repetitionPenalty": 1.1,
+      "repetitionContext": 20,
+      "maxTokens": 8192
+    },
+    "reasoning": {
+      "enableThinking": true,
+      "thinkingBudget": 256
+    },
+    "profiles": {
+      "thinking-general": {
+        "sampling": {
+          "temperature": 0.8,
+          "topP": 0.95,
+          "topK": 64,
+          "repetitionPenalty": 1.1,
+          "repetitionContext": 20,
+          "maxTokens": 2048
+        }
+      },
+      "thinking-coding": {
+        "sampling": {
+          "temperature": 0.6,
+          "topP": 0.9,
+          "topK": 40,
+          "repetitionPenalty": 1.05,
+          "repetitionContext": 32,
+          "maxTokens": 2048
+        }
+      },
+      "thinking-precise": {
+        "sampling": {
+          "temperature": 0.3,
+          "topP": 0.9,
+          "topK": 20,
+          "repetitionPenalty": 1.05,
+          "repetitionContext": 20,
+          "maxTokens": 4096,
+          "seed": 0
+        }
+      },
+      "instruct": {
+        "sampling": {
+          "temperature": 0.7,
+          "topP": 0.95,
+          "topK": 64,
+          "repetitionPenalty": 1.1,
+          "repetitionContext": 20,
+          "maxTokens": 4096
+        }
+      },
+      "creative": {
+        "sampling": {
+          "temperature": 1.1,
+          "topP": 0.95,
+          "topK": 64,
+          "repetitionPenalty": 1,
+          "repetitionContext": 20,
+          "maxTokens": 8192
+        }
+      }
+    }
+  },
+  "evalHints": {
+    "sniffThresholds": {
+      "inlineJsMinBytes": 1000
+    }
+  },
+  "llamaCpp": {
+    "huggingfaceRepo": "ggml-org/gemma-4-12B-it-GGUF",
+    "filename": "gemma-4-12B-it-Q8_0.gguf",
+    "quantization": "Q8_0",
+    "residentBytes": 15761707354,
+    "draftModelFilename": "mtp-gemma-4-12B-it-Q8_0.gguf"
+  },
+  "releasePatch": {
+    "tuning": {
+      "engine": {
+        "llamaCpp": {
+          "spec": {
+            "mtp": true,
+            "nMax": 4
+          }
+        }
+      }
+    }
+  },
+  "mlx": {
+    "huggingfaceRepo": "mlx-community/gemma-4-12b-it-8bit",
+    "quantization": "8bit"
+  }
+}

package/authoring/chat-models/gemma4-26b-q4.json

@@ -0,0 +1,90 @@
+{
+  "id": "gemma4-26b-q4",
+  "name": "Gemma 4 (26B, MoE, Q4)",
+  "description": "Google's Gemma 4 mixture-of-experts model — 25.2B total parameters with 3.8B active per token. Multimodal with long-context agentic workflows. The full 26B has to fit in memory but only ~4B parameters fire per token, so on capable hardware it runs noticeably faster than the dense 31B at a similar memory cost — the recommended high-end on-device default.",
+  "tags": [
+    "google",
+    "multimodal",
+    "vision",
+    "tools",
+    "mix of experts",
+    "agentic",
+    "qat"
+  ],
+  "category": "general",
+  "maintainer": {
+    "name": "Google",
+    "url": "https://huggingface.co/google/gemma-4-26B-A4B-it"
+  },
+  "version": "1.2.1",
+  "updatedAt": "2026-07-28T00:00:00Z",
+  "license": "Apache-2.0",
+  "recoScore": 15,
+  "parameterSize": "25.2B",
+  "approxSizeBytes": 14249047104,
+  "supportsTools": true,
+  "contextWindow": 256000,
+  "upstream": "https://huggingface.co/google/gemma-4-26B-A4B-it",
+  "style": {
+    "family": "gemma",
+    "reasoningFormat": "channel",
+    "toolCallFormat": "function-call"
+  },
+  "behaviors": [
+    "reasoning.strip-channel-tags",
+    "reasoning.capture-pre-tool-prose",
+    "prompt.tool-cookbook-full",
+    "prompt.meester-build-prelude",
+    "fabrication.detect-past-tense-no-tools",
+    "fabrication.detect-claim-without-tool",
+    "turn.preamble-folding",
+    {
+      "id": "turn.ramble-detection",
+      "config": {
+        "coldThreshold": 8000,
+        "postActionThreshold": 2000
+      }
+    },
+    "turn.auto-acknowledge-tool-errors",
+    {
+      "id": "turn.continuation-budget",
+      "config": {
+        "count": 4
+      }
+    },
+    "mcp.validate-ids-strict",
+    {
+      "id": "turn.single-tool-per-turn",
+      "config": {
+        "meesterOnly": true
+      }
+    },
+    "parse.gemma-special-token"
+  ],
+  "ollama": {
+    "tag": "gemma4:26b-a4b-it-qat"
+  },
+  "llamaCpp": {
+    "huggingfaceRepo": "unsloth/gemma-4-26B-A4B-it-qat-GGUF",
+    "filename": "gemma-4-26B-A4B-it-qat-UD-Q4_K_XL.gguf",
+    "quantization": "UD-Q4_K_XL",
+    "residentBytes": 17401183718,
+    "draftModelFilename": "MTP/mtp-gemma-4-26B-A4B-it-Q4_0.gguf"
+  },
+  "releasePatch": {
+    "tuning": {
+      "engine": {
+        "llamaCpp": {
+          "spec": {
+            "mtp": true,
+            "nMax": 4
+          }
+        }
+      }
+    }
+  },
+  "mlx": {
+    "huggingfaceRepo": "mlx-community/gemma-4-26B-A4B-it-qat-4bit",
+    "quantization": "4bit"
+  }
+}