@bendyline/gezel-inference 1.0.3 → 1.0.5

package/dist/catalog/chat-models/na/nanbeige4.2-3b-q4/manifest.json

@@ -3,7 +3,7 @@
   "kind": "chat-model",
   "id": "nanbeige4.2-3b-q4",
   "name": "Nanbeige 4.2 (3B, Q4)",
-  "description": "Nanbeige LLM Lab's compact agentic reasoning model. A looped transformer — 22 layers run twice over shared weights — so a 3B-parameter download does the compute of a much deeper stack. Text-only, with native thinking, XML tool calling, English + Chinese, and a 256K context window. Q4 builds suit machines with 8GB+ usable memory.",
+  "description": "Nanbeige LLM Lab's compact agentic reasoning model. Its 3B label counts non-embedding parameters; total parameters are about 4B. The 22 transformer layers run twice over shared weights, with separate context caches. The Q4 download is 2.5 GiB, but one chat's FP16 context cache alone needs 5.5 GiB at 32K tokens or 11 GiB at 64K, plus model weights and engine overhead. Allow roughly 10 GiB of free GPU memory for a 32K, single-chat FP16 setup; longer context or more simultaneous chats needs substantially more headroom. Text-only, with native thinking, XML tool calling, English + Chinese, and support for up to 256K context when memory permits.",
   "tags": [
     "nanbeige",
     "reasoning",

package/dist/catalog/chat-models/ne/nemotron3.5-bolt-e4b-q4/manifest.json

@@ -0,0 +1,142 @@
+{
+  "schemaVersion": 1,
+  "kind": "chat-model",
+  "id": "nemotron3.5-bolt-e4b-q4",
+  "name": "Nemotron 3.5 Bolt (E4B, Q4)",
+  "description": "NVIDIA's conversational model built on Gemma 4 E4B, with native tool calling, optional reasoning, and a 64K context window. Text only. The official Q4_K_M download is about 6.2GB; working memory depends on context and hardware.",
+  "tags": [
+    "nvidia",
+    "conversation",
+    "tools",
+    "reasoning",
+    "open-weights"
+  ],
+  "category": "general",
+  "maintainer": {
+    "name": "NVIDIA",
+    "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Bolt-E4B-4GB"
+  },
+  "version": "1.0.1",
+  "updatedAt": "2026-10-25T00:00:00Z",
+  "license": "OpenMDW-1.1",
+  "licenseClass": "open",
+  "licenseShortName": "OpenMDW 1.1",
+  "licenseUrl": "https://openmdw.ai/license/1-1/",
+  "parameterSize": "8B",
+  "approxSizeBytes": 6180446208,
+  "supportsTools": true,
+  "contextWindow": 65536,
+  "upstream": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Bolt-E4B-4GB",
+  "style": {
+    "family": "gemma",
+    "reasoningFormat": "channel",
+    "toolCallFormat": "function-call"
+  },
+  "behaviors": [
+    "reasoning.strip-channel-tags",
+    "prompt.private-reasoning-guidance",
+    "prompt.tool-cookbook-condensed",
+    "prompt.meester-build-prelude",
+    "fabrication.detect-past-tense-no-tools",
+    "fabrication.detect-claim-without-tool",
+    "turn.auto-acknowledge-tool-errors",
+    "mcp.compact-tool-schemas",
+    "mcp.relax-required-fields",
+    "mcp.default-missing-fields",
+    "turn.preamble-folding",
+    "turn.ramble-detection",
+    "parse.gemma-special-token",
+    "prompt.prefer-writefile-edits",
+    "prompt.derive-by-execution"
+  ],
+  "tuning": {
+    "sampling": {
+      "temperature": 1,
+      "topP": 0.95,
+      "topK": 64,
+      "repetitionPenalty": 1.1,
+      "repetitionContext": 20,
+      "maxTokens": 8192
+    },
+    "reasoning": {
+      "enableThinking": true,
+      "thinkingBudget": 2048
+    },
+    "engine": {
+      "llamaCpp": {
+        "spec": {
+          "type": "none",
+          "mtp": false
+        },
+        "contextSize": 65536
+      }
+    },
+    "profiles": {
+      "thinking-general": {
+        "sampling": {
+          "temperature": 0.9,
+          "topP": 0.95,
+          "topK": 64,
+          "repetitionPenalty": 1.1,
+          "repetitionContext": 20,
+          "maxTokens": 8192
+        }
+      },
+      "thinking-coding": {
+        "sampling": {
+          "temperature": 0.3,
+          "topP": 0.9,
+          "topK": 40,
+          "repetitionPenalty": 1.1,
+          "repetitionContext": 32,
+          "maxTokens": 12288
+        }
+      },
+      "thinking-precise": {
+        "sampling": {
+          "temperature": 0.3,
+          "topP": 0.9,
+          "topK": 20,
+          "repetitionPenalty": 1.1,
+          "repetitionContext": 20,
+          "maxTokens": 4096,
+          "seed": 0
+        }
+      },
+      "instruct": {
+        "sampling": {
+          "temperature": 0.8,
+          "topP": 0.9,
+          "topK": 40,
+          "repetitionPenalty": 1.1,
+          "repetitionContext": 20,
+          "maxTokens": 4096
+        },
+        "reasoning": {
+          "enableThinking": false
+        }
+      },
+      "creative": {
+        "sampling": {
+          "temperature": 1,
+          "topP": 0.95,
+          "topK": 80,
+          "repetitionPenalty": 1.1,
+          "repetitionContext": 20,
+          "maxTokens": 8192
+        },
+        "reasoning": {
+          "enableThinking": false
+        }
+      }
+    }
+  },
+  "llamaCpp": {
+    "huggingfaceRepo": "nvidia/NVIDIA-Nemotron-3.5-Bolt-E4B-4GB",
+    "revision": "1db54d5b12533ff1bcaa8e42cf4c667988a1f1be",
+    "filename": "Nemotron-3.5-Bolt-E4B-Q4_K_M.gguf",
+    "sha256": "c8af2b8aea10a690a1d9491db744a9446660173cb38fc489612b556615a69612",
+    "approxSizeBytes": 6180446208,
+    "quantization": "Q4_K_M"
+  }
+}

package/dist/catalog/chat-models/ne/nemotron3.5-bolt-e4b-q4/versions/1.0.0/manifest.json

@@ -0,0 +1,14 @@
+{
+  "schemaVersion": 1,
+  "version": "1.0.0",
+  "releasedAt": "2026-10-23T00:00:00Z",
+  "approxSizeBytes": 6180446208,
+  "llamaCpp": {
+    "huggingfaceRepo": "nvidia/NVIDIA-Nemotron-3.5-Bolt-E4B-4GB",
+    "revision": "fc2c9fed2eaabca56d44a5be74a448b14d4bc183",
+    "filename": "Nemotron-3.5-Bolt-E4B-Q4_K_M.gguf",
+    "sha256": "5442bfb3b1384ff3774c9989e9457dad81bbb5830fbcf96adfdde40078831d5a",
+    "approxSizeBytes": 6180446208,
+    "quantization": "Q4_K_M"
+  }
+}

package/dist/catalog/chat-models/ne/nemotron3.5-bolt-e4b-q4/versions/1.0.1/manifest.json

@@ -0,0 +1,14 @@
+{
+  "schemaVersion": 1,
+  "version": "1.0.1",
+  "releasedAt": "2026-10-25T00:00:00Z",
+  "approxSizeBytes": 6180446208,
+  "llamaCpp": {
+    "huggingfaceRepo": "nvidia/NVIDIA-Nemotron-3.5-Bolt-E4B-4GB",
+    "revision": "1db54d5b12533ff1bcaa8e42cf4c667988a1f1be",
+    "filename": "Nemotron-3.5-Bolt-E4B-Q4_K_M.gguf",
+    "sha256": "c8af2b8aea10a690a1d9491db744a9446660173cb38fc489612b556615a69612",
+    "approxSizeBytes": 6180446208,
+    "quantization": "Q4_K_M"
+  }
+}