@bendyline/gezel-inference 1.0.1 → 1.0.3

package/dist/catalog/chat-models/de/deepseek-v4.1-flash-552b-q2/manifest.json

@@ -0,0 +1,120 @@
+{
+  "schemaVersion": 1,
+  "kind": "chat-model",
+  "id": "deepseek-v4.1-flash-552b-q2",
+  "name": "DeepSeek V4.1 Flash (IQ2_XXS)",
+  "description": "antirez's 2-bit GGUF of DeepSeek V4.1 Flash 1017, the October 2026 update to V4.1 Flash with large gains in reasoning, coding and agentic work. It is a 552B-parameter Mixture-of-Experts (8B active per token while reading a prompt, 16B while writing) with a 1M-token context and 196B parameters of Engram memory looked up from disk. The file is 341 GiB: 152 GiB of model weights plus 189 GiB of Engram tables that are never loaded into memory, so it needs a fast local SSD with room to spare. On a 128 GB Mac or DGX Spark the routed experts stream from the SSD, which makes it slower than models that fit in memory. Text only in Gezel for now. Produced for antirez's `ds4` inference engine.",
+  "tags": [
+    "deepseek",
+    "reasoning",
+    "agentic",
+    "moe",
+    "tools",
+    "2-bit",
+    "iq2",
+    "long-context",
+    "large"
+  ],
+  "category": "reasoning",
+  "maintainer": {
+    "name": "antirez",
+    "url": "https://huggingface.co/antirez/deepseek-v4.1-flash-gguf"
+  },
+  "maker": {
+    "name": "DeepSeek",
+    "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017"
+  },
+  "version": "1.0.0",
+  "updatedAt": "2026-10-19T00:00:00Z",
+  "license": "MIT",
+  "licenseClass": "open",
+  "licenseShortName": "MIT",
+  "parameterSize": "552B",
+  "approxSizeBytes": 365713686528,
+  "supportsTools": true,
+  "contextWindow": 1048576,
+  "upstream": "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017",
+  "style": {
+    "family": "deepseek",
+    "reasoningFormat": "think",
+    "toolCallFormat": "function-call"
+  },
+  "behaviors": [
+    "reasoning.strip-think-tags",
+    "prompt.private-reasoning-guidance",
+    "prompt.tool-cookbook-condensed",
+    "prompt.meester-build-prelude",
+    "mcp.compact-tool-schemas",
+    "fabrication.detect-past-tense-no-tools",
+    "turn.preamble-folding"
+  ],
+  "tuning": {
+    "sampling": {
+      "temperature": 1,
+      "topP": 1,
+      "maxTokens": 16384
+    },
+    "reasoning": {
+      "enableThinking": true,
+      "thinkingBudget": 4096
+    },
+    "profiles": {
+      "thinking-general": {
+        "sampling": {
+          "temperature": 1,
+          "topP": 1,
+          "maxTokens": 16384
+        }
+      },
+      "thinking-coding": {
+        "sampling": {
+          "temperature": 1,
+          "topP": 1,
+          "maxTokens": 16384
+        }
+      },
+      "thinking-precise": {
+        "sampling": {
+          "temperature": 1,
+          "topP": 1,
+          "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
+        }
+      }
+    }
+  },
+  "ds4": {
+    "huggingfaceRepo": "antirez/deepseek-v4.1-flash-gguf",
+    "revision": "9dcee51692e6540f697141fc98b097e3c96492a3",
+    "filename": "DeepSeek-V4.1-Flash-1017-IQ2_XXS-Q2_K-imatrix.gguf",
+    "sha256": "5296e85ff64b42ecb62c8d0c0c54e1caa07e3df8e93112f2365ebbaed62ce93a",
+    "approxSizeBytes": 365713686528,
+    "quantization": "IQ2_XXS/Q2_K",
+    "residentBytes": 88315265024,
+    "residentWeightBytes": 162955640832,
+    "kvBytesPerToken": 11469,
+    "residentCtxTokens": 131072,
+    "cacheExpertsBytes": 68719476736,
+    "ssdStreaming": true
+  }
+}

package/dist/catalog/chat-models/de/deepseek-v4.1-flash-552b-q2/versions/1.0.0/manifest.json

@@ -0,0 +1,21 @@
+{
+  "schemaVersion": 1,
+  "version": "1.0.0",
+  "releasedAt": "2026-10-19T00:00:00Z",
+  "minGezelVersion": "1.26261",
+  "approxSizeBytes": 365713686528,
+  "ds4": {
+    "huggingfaceRepo": "antirez/deepseek-v4.1-flash-gguf",
+    "revision": "9dcee51692e6540f697141fc98b097e3c96492a3",
+    "filename": "DeepSeek-V4.1-Flash-1017-IQ2_XXS-Q2_K-imatrix.gguf",
+    "sha256": "5296e85ff64b42ecb62c8d0c0c54e1caa07e3df8e93112f2365ebbaed62ce93a",
+    "approxSizeBytes": 365713686528,
+    "quantization": "IQ2_XXS/Q2_K",
+    "residentBytes": 88315265024,
+    "residentWeightBytes": 162955640832,
+    "kvBytesPerToken": 11469,
+    "residentCtxTokens": 131072,
+    "cacheExpertsBytes": 68719476736,
+    "ssdStreaming": true
+  }
+}

package/dist/catalog/chat-models/gl/glm-5.3-flash-320b-q2/manifest.json

@@ -28,8 +28,8 @@
     "name": "Z.ai",
     "url": "https://huggingface.co/zai-org/GLM-5.3-Flash"
   },
-  "version": "1.0.1",
-  "updatedAt": "2026-09-06T00:00:00Z",
+  "version": "1.0.3",
+  "updatedAt": "2026-10-20T00:00:00Z",
   "license": "MIT",
   "licenseClass": "open",
   "licenseShortName": "MIT",
@@ -130,9 +130,10 @@
       "sha256": "ae23e14c6979e889051b2e4a39351abcdafb161e18e606fae4d8c40095a4bf3a",
       "sizeBytes": 1127280960
     },
-    "residentBytes": 100865554432,
-    "ssdStreaming": false,
-    "maxLaunchCtx": 32768,
+    "residentBytes": 47244640256,
+    "cacheExpertsBytes": 34359738368,
+    "ssdStreaming": true,
+    "maxLaunchCtx": 65536,
     "mtp": {
       "exactSampling": true
     }

package/dist/catalog/chat-models/gl/glm-5.3-flash-320b-q2/versions/1.0.2/manifest.json

@@ -0,0 +1,27 @@
+{
+  "schemaVersion": 1,
+  "version": "1.0.2",
+  "releasedAt": "2026-10-17T00:00:00Z",
+  "minGezelVersion": "1.26248",
+  "approxSizeBytes": 97633097344,
+  "ds4": {
+    "huggingfaceRepo": "antirez/glm-5.3-flash-gguf",
+    "revision": "b2fa29d7a6b410db11221c904973967b80b760f5",
+    "filename": "GLM-5.3-Flash-Q2.gguf",
+    "sha256": "e81fd6241c6e55a64e1e14e47a3eab61a173fa8d7e4b5c1d1848827119705b32",
+    "approxSizeBytes": 96505816384,
+    "quantization": "IQ2_XXS/Q2_K",
+    "visionEncoder": {
+      "filename": "GLM-5.3-Flash-Vision-Encoder.gguf",
+      "sha256": "ae23e14c6979e889051b2e4a39351abcdafb161e18e606fae4d8c40095a4bf3a",
+      "sizeBytes": 1127280960
+    },
+    "residentBytes": 47244640256,
+    "cacheExpertsBytes": 34359738368,
+    "ssdStreaming": true,
+    "maxLaunchCtx": 32768,
+    "mtp": {
+      "exactSampling": true
+    }
+  }
+}

package/dist/catalog/chat-models/gl/glm-5.3-flash-320b-q2/versions/1.0.3/manifest.json

@@ -0,0 +1,27 @@
+{
+  "schemaVersion": 1,
+  "version": "1.0.3",
+  "releasedAt": "2026-10-20T00:00:00Z",
+  "minGezelVersion": "1.26248",
+  "approxSizeBytes": 97633097344,
+  "ds4": {
+    "huggingfaceRepo": "antirez/glm-5.3-flash-gguf",
+    "revision": "b2fa29d7a6b410db11221c904973967b80b760f5",
+    "filename": "GLM-5.3-Flash-Q2.gguf",
+    "sha256": "e81fd6241c6e55a64e1e14e47a3eab61a173fa8d7e4b5c1d1848827119705b32",
+    "approxSizeBytes": 96505816384,
+    "quantization": "IQ2_XXS/Q2_K",
+    "visionEncoder": {
+      "filename": "GLM-5.3-Flash-Vision-Encoder.gguf",
+      "sha256": "ae23e14c6979e889051b2e4a39351abcdafb161e18e606fae4d8c40095a4bf3a",
+      "sizeBytes": 1127280960
+    },
+    "residentBytes": 47244640256,
+    "cacheExpertsBytes": 34359738368,
+    "ssdStreaming": true,
+    "maxLaunchCtx": 65536,
+    "mtp": {
+      "exactSampling": true
+    }
+  }
+}