[Sync] llm stuff

2026-09-17 20:58:21 +02:00
parent ed7e34552d
commit d8f6b30e2c
52 changed files with 2233 additions and 237 deletions
@@ -12,7 +12,7 @@
      "models": [
        {
          "id": "gemma-4-E4B-it-UD-Q8_K_XL",
-          "name": "Gemma 4 E4B · 64k · vision — fast generalist, long docs",
+          "name": "Gemma 4 E4B · 64k · vision — fast generalist, long docs (~57 t/s)",
          "reasoning": false,
          "input": ["text", "image"],
          "contextWindow": 65536,
@@ -21,7 +21,7 @@
        },
        {
          "id": "gemma-4-26B-A4B-it-UD-IQ4_XS",
-          "name": "Gemma 4 26B · 24k · vision — quality generalist",
+          "name": "Gemma 4 26B · 24k · vision — quality generalist (~34 t/s)",
          "reasoning": false,
          "input": ["text", "image"],
          "contextWindow": 24576,
@@ -30,7 +30,7 @@
        },
        {
          "id": "Qwen3-Coder-30B-Instruct-UD-Q3_K_XL",
-          "name": "Qwen3 Coder 30B · 32k — main agent coder",
+          "name": "Qwen3 Coder 30B · 32k — main agent coder (~30 t/s)",
          "reasoning": false,
          "input": ["text"],
          "contextWindow": 32768,
@@ -39,7 +39,7 @@
        },
        {
          "id": "Qwen3-Coder-Next-UD-IQ3_XXS",
-          "name": "Qwen3 Coder Next 80B · 128k — long sessions",
+          "name": "Qwen3 Coder Next 80B · 128k — long sessions (~16 t/s)",
          "reasoning": false,
          "input": ["text"],
          "contextWindow": 131072,
@@ -48,7 +48,7 @@
        },
        {
          "id": "Qwen3.6-35B-A3B-MTP-UD-IQ3_XXS",
-          "name": "Qwen3.6 35B · 24k · vision — daily driver",
+          "name": "Qwen3.6 35B · 24k · vision — daily driver (~36 t/s)",
          "reasoning": false,
          "input": ["text", "image"],
          "contextWindow": 24576,
@@ -57,7 +57,7 @@
        },
        {
          "id": "Qwen3.6-35B-A3B-Thinking",
-          "name": "Qwen3.6 35B Thinking · 24k · vision — hard problems",
+          "name": "Qwen3.6 35B Thinking · 24k · vision — hard problems (~39 t/s)",
          "reasoning": true,
          "compat": { "thinkingFormat": "qwen-chat-template" },
          "input": ["text", "image"],
@@ -67,7 +67,7 @@
        },
        {
          "id": "Qwen3.5-9B-UD-Q6_K_XL",
-          "name": "Qwen3.5 9B · 32k · vision — quick tasks",
+          "name": "Qwen3.5 9B · 32k · vision — quick tasks (~32 t/s)",
          "reasoning": true,
          "compat": { "thinkingFormat": "qwen-chat-template" },
          "input": ["text", "image"],
@@ -77,17 +77,17 @@
        },
        {
          "id": "Qwen3.8-27B-UD-IQ3_XXS",
-          "name": "Qwen3.8 27B · 24k · vision — hybrid reasoner",
+          "name": "Qwen3.8 27B · 24k · vision — hybrid reasoner, quiet desktop only (~15 t/s)",
          "reasoning": true,
          "compat": { "thinkingFormat": "qwen-chat-template" },
          "input": ["text", "image"],
          "contextWindow": 24576,
-          "maxTokens": 8192,
+          "maxTokens": 4096,
          "cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 }
        },
        {
          "id": "GLM-4.7-Flash-UD-Q4_K_XL",
-          "name": "GLM-4.7 Flash · 24k — quality coder",
+          "name": "GLM-4.7 Flash · 24k — quality coder (~21 t/s)",
          "reasoning": true,
          "compat": { "thinkingFormat": "qwen-chat-template" },
          "input": ["text"],
@@ -97,7 +97,7 @@
        },
        {
          "id": "gpt-oss-20b",
-          "name": "gpt-oss 20B · 64k — fast reasoning + tools",
+          "name": "gpt-oss 20B · 64k — fast reasoning + tools (~38 t/s)",
          "reasoning": true,
          "input": ["text"],
          "contextWindow": 65536,
@@ -106,7 +106,7 @@
        },
        {
          "id": "gpt-oss-20b-low",
-          "name": "gpt-oss 20B low · 64k — snappy answers",
+          "name": "gpt-oss 20B low · 64k — snappy answers (~37 t/s)",
          "reasoning": true,
          "input": ["text"],
          "contextWindow": 65536,
@@ -127,7 +127,7 @@
      "models": [
        {
          "id": "gemma-4-E4B-it-UD-Q8_K_XL",
-          "name": "Gemma 4 E4B · 64k · vision — fast generalist, long docs",
+          "name": "Gemma 4 E4B · 64k · vision — fast generalist, long docs (~57 t/s)",
          "reasoning": false,
          "input": ["text", "image"],
          "contextWindow": 65536,
@@ -136,7 +136,7 @@
        },
        {
          "id": "gemma-4-26B-A4B-it-UD-IQ4_XS",
-          "name": "Gemma 4 26B · 24k · vision — quality generalist",
+          "name": "Gemma 4 26B · 24k · vision — quality generalist (~34 t/s)",
          "reasoning": false,
          "input": ["text", "image"],
          "contextWindow": 24576,
@@ -145,7 +145,7 @@
        },
        {
          "id": "Qwen3-Coder-30B-Instruct-UD-Q3_K_XL",
-          "name": "Qwen3 Coder 30B · 32k — main agent coder",
+          "name": "Qwen3 Coder 30B · 32k — main agent coder (~30 t/s)",
          "reasoning": false,
          "input": ["text"],
          "contextWindow": 32768,
@@ -154,7 +154,7 @@
        },
        {
          "id": "Qwen3-Coder-Next-UD-IQ3_XXS",
-          "name": "Qwen3 Coder Next 80B · 128k — long sessions",
+          "name": "Qwen3 Coder Next 80B · 128k — long sessions (~16 t/s)",
          "reasoning": false,
          "input": ["text"],
          "contextWindow": 131072,
@@ -163,7 +163,7 @@
        },
        {
          "id": "Qwen3.6-35B-A3B-MTP-UD-IQ3_XXS",
-          "name": "Qwen3.6 35B · 24k · vision — daily driver",
+          "name": "Qwen3.6 35B · 24k · vision — daily driver (~36 t/s)",
          "reasoning": false,
          "input": ["text", "image"],
          "contextWindow": 24576,
@@ -172,7 +172,7 @@
        },
        {
          "id": "Qwen3.6-35B-A3B-Thinking",
-          "name": "Qwen3.6 35B Thinking · 24k · vision — hard problems",
+          "name": "Qwen3.6 35B Thinking · 24k · vision — hard problems (~39 t/s)",
          "reasoning": true,
          "compat": { "thinkingFormat": "qwen-chat-template" },
          "input": ["text", "image"],
@@ -182,7 +182,7 @@
        },
        {
          "id": "Qwen3.5-9B-UD-Q6_K_XL",
-          "name": "Qwen3.5 9B · 32k · vision — quick tasks",
+          "name": "Qwen3.5 9B · 32k · vision — quick tasks (~32 t/s)",
          "reasoning": true,
          "compat": { "thinkingFormat": "qwen-chat-template" },
          "input": ["text", "image"],
@@ -192,17 +192,17 @@
        },
        {
          "id": "Qwen3.8-27B-UD-IQ3_XXS",
-          "name": "Qwen3.8 27B · 24k · vision — hybrid reasoner",
+          "name": "Qwen3.8 27B · 24k · vision — hybrid reasoner, quiet desktop only (~15 t/s)",
          "reasoning": true,
          "compat": { "thinkingFormat": "qwen-chat-template" },
          "input": ["text", "image"],
          "contextWindow": 24576,
-          "maxTokens": 8192,
+          "maxTokens": 4096,
          "cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 }
        },
        {
          "id": "GLM-4.7-Flash-UD-Q4_K_XL",
-          "name": "GLM-4.7 Flash · 24k — quality coder",
+          "name": "GLM-4.7 Flash · 24k — quality coder (~21 t/s)",
          "reasoning": true,
          "compat": { "thinkingFormat": "qwen-chat-template" },
          "input": ["text"],
@@ -212,7 +212,7 @@
        },
        {
          "id": "gpt-oss-20b",
-          "name": "gpt-oss 20B · 64k — fast reasoning + tools",
+          "name": "gpt-oss 20B · 64k — fast reasoning + tools (~38 t/s)",
          "reasoning": true,
          "input": ["text"],
          "contextWindow": 65536,
@@ -221,7 +221,7 @@
        },
        {
          "id": "gpt-oss-20b-low",
-          "name": "gpt-oss 20B low · 64k — snappy answers",
+          "name": "gpt-oss 20B low · 64k — snappy answers (~37 t/s)",
          "reasoning": true,
          "input": ["text"],
          "contextWindow": 65536,