[Update] bulk update
This commit is contained in:
@@ -61,6 +61,9 @@ clients:
|
||||
- name: Qwen3.5-9B-UD-Q6_K_XL
|
||||
max_input_tokens: 30000 # ctx 32768
|
||||
supports_vision: true
|
||||
- name: Qwen3.8-27B-UD-IQ3_XXS
|
||||
max_input_tokens: 22000 # ctx 24576
|
||||
supports_vision: true
|
||||
- name: gemma-4-26B-A4B-it-UD-IQ4_XS
|
||||
max_input_tokens: 22000 # ctx 24576
|
||||
supports_vision: true
|
||||
|
||||
@@ -158,6 +158,7 @@
|
||||
"Qwen3.6-35B-A3B-MTP-UD-IQ3_XXS": { "name": "Qwen3.6 35B · 24k · vision — daily driver (remote)", "attachment": true, "limit": { "context": 24576, "output": 8192 } },
|
||||
"Qwen3.6-35B-A3B-Thinking": { "name": "Qwen3.6 35B Thinking · 24k · vision — hard problems (remote)", "attachment": true, "limit": { "context": 24576, "output": 8192 } },
|
||||
"Qwen3.5-9B-UD-Q6_K_XL": { "name": "Qwen3.5 9B · 32k · vision — quick tasks (remote)", "attachment": true, "limit": { "context": 32768, "output": 8192 } },
|
||||
"Qwen3.8-27B-UD-IQ3_XXS": { "name": "Qwen3.8 27B · 24k · vision — hybrid reasoner (remote)", "attachment": true, "limit": { "context": 24576, "output": 8192 } },
|
||||
"gemma-4-26B-A4B-it-UD-IQ4_XS": { "name": "Gemma 4 26B · 24k · vision — quality generalist (remote)", "attachment": true, "limit": { "context": 24576, "output": 4096 } },
|
||||
"gemma-4-E4B-it-UD-Q8_K_XL": { "name": "Gemma 4 E4B · 64k · vision — fast generalist, long docs (remote)", "attachment": true, "limit": { "context": 65536, "output": 16384 } },
|
||||
"GLM-4.7-Flash-UD-Q4_K_XL": { "name": "GLM-4.7 Flash · 24k — quality coder (remote)", "limit": { "context": 24576, "output": 8192 } },
|
||||
@@ -177,6 +178,7 @@
|
||||
"Qwen3.6-35B-A3B-MTP-UD-IQ3_XXS": { "name": "Qwen3.6 35B · 24k · vision — daily driver", "attachment": true, "limit": { "context": 24576, "output": 8192 } },
|
||||
"Qwen3.6-35B-A3B-Thinking": { "name": "Qwen3.6 35B Thinking · 24k · vision — hard problems", "attachment": true, "limit": { "context": 24576, "output": 8192 } },
|
||||
"Qwen3.5-9B-UD-Q6_K_XL": { "name": "Qwen3.5 9B · 32k · vision — quick tasks", "attachment": true, "limit": { "context": 32768, "output": 8192 } },
|
||||
"Qwen3.8-27B-UD-IQ3_XXS": { "name": "Qwen3.8 27B · 24k · vision — hybrid reasoner", "attachment": true, "limit": { "context": 24576, "output": 8192 } },
|
||||
"gemma-4-26B-A4B-it-UD-IQ4_XS": { "name": "Gemma 4 26B · 24k · vision — quality generalist", "attachment": true, "limit": { "context": 24576, "output": 4096 } },
|
||||
"gemma-4-E4B-it-UD-Q8_K_XL": { "name": "Gemma 4 E4B · 64k · vision — fast generalist, long docs", "attachment": true, "limit": { "context": 65536, "output": 16384 } },
|
||||
"GLM-4.7-Flash-UD-Q4_K_XL": { "name": "GLM-4.7 Flash · 24k — quality coder", "limit": { "context": 24576, "output": 8192 } },
|
||||
|
||||
+40
-20
@@ -101,7 +101,7 @@
|
||||
"reasoning": false,
|
||||
"input": ["text", "image"],
|
||||
"contextWindow": 65536,
|
||||
"maxTokens": 16384,
|
||||
"maxTokens": 2048,
|
||||
"cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 }
|
||||
},
|
||||
{
|
||||
@@ -110,7 +110,7 @@
|
||||
"reasoning": false,
|
||||
"input": ["text", "image"],
|
||||
"contextWindow": 24576,
|
||||
"maxTokens": 4096,
|
||||
"maxTokens": 2048,
|
||||
"cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 }
|
||||
},
|
||||
{
|
||||
@@ -119,7 +119,7 @@
|
||||
"reasoning": false,
|
||||
"input": ["text"],
|
||||
"contextWindow": 32768,
|
||||
"maxTokens": 4096,
|
||||
"maxTokens": 2048,
|
||||
"cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 }
|
||||
},
|
||||
{
|
||||
@@ -128,7 +128,7 @@
|
||||
"reasoning": false,
|
||||
"input": ["text"],
|
||||
"contextWindow": 131072,
|
||||
"maxTokens": 16384,
|
||||
"maxTokens": 2048,
|
||||
"cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 }
|
||||
},
|
||||
{
|
||||
@@ -137,7 +137,7 @@
|
||||
"reasoning": false,
|
||||
"input": ["text", "image"],
|
||||
"contextWindow": 24576,
|
||||
"maxTokens": 8192,
|
||||
"maxTokens": 2048,
|
||||
"cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 }
|
||||
},
|
||||
{
|
||||
@@ -147,7 +147,7 @@
|
||||
"compat": { "thinkingFormat": "qwen-chat-template" },
|
||||
"input": ["text", "image"],
|
||||
"contextWindow": 24576,
|
||||
"maxTokens": 8192,
|
||||
"maxTokens": 2048,
|
||||
"cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 }
|
||||
},
|
||||
{
|
||||
@@ -157,7 +157,17 @@
|
||||
"compat": { "thinkingFormat": "qwen-chat-template" },
|
||||
"input": ["text", "image"],
|
||||
"contextWindow": 32768,
|
||||
"maxTokens": 8192,
|
||||
"maxTokens": 2048,
|
||||
"cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 }
|
||||
},
|
||||
{
|
||||
"id": "Qwen3.8-27B-UD-IQ3_XXS",
|
||||
"name": "Qwen3.8 27B · 24k · vision — hybrid reasoner",
|
||||
"reasoning": true,
|
||||
"compat": { "thinkingFormat": "qwen-chat-template" },
|
||||
"input": ["text", "image"],
|
||||
"contextWindow": 24576,
|
||||
"maxTokens": 2048,
|
||||
"cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 }
|
||||
},
|
||||
{
|
||||
@@ -167,7 +177,7 @@
|
||||
"compat": { "thinkingFormat": "qwen-chat-template" },
|
||||
"input": ["text"],
|
||||
"contextWindow": 24576,
|
||||
"maxTokens": 8192,
|
||||
"maxTokens": 2048,
|
||||
"cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 }
|
||||
},
|
||||
{
|
||||
@@ -176,7 +186,7 @@
|
||||
"reasoning": true,
|
||||
"input": ["text"],
|
||||
"contextWindow": 65536,
|
||||
"maxTokens": 8192,
|
||||
"maxTokens": 2048,
|
||||
"cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 }
|
||||
},
|
||||
{
|
||||
@@ -185,7 +195,7 @@
|
||||
"reasoning": true,
|
||||
"input": ["text"],
|
||||
"contextWindow": 65536,
|
||||
"maxTokens": 8192,
|
||||
"maxTokens": 2048,
|
||||
"cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 }
|
||||
}
|
||||
]
|
||||
@@ -206,7 +216,7 @@
|
||||
"reasoning": false,
|
||||
"input": ["text", "image"],
|
||||
"contextWindow": 65536,
|
||||
"maxTokens": 16384,
|
||||
"maxTokens": 2048,
|
||||
"cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 }
|
||||
},
|
||||
{
|
||||
@@ -215,7 +225,7 @@
|
||||
"reasoning": false,
|
||||
"input": ["text", "image"],
|
||||
"contextWindow": 24576,
|
||||
"maxTokens": 4096,
|
||||
"maxTokens": 2048,
|
||||
"cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 }
|
||||
},
|
||||
{
|
||||
@@ -224,7 +234,7 @@
|
||||
"reasoning": false,
|
||||
"input": ["text"],
|
||||
"contextWindow": 32768,
|
||||
"maxTokens": 4096,
|
||||
"maxTokens": 2048,
|
||||
"cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 }
|
||||
},
|
||||
{
|
||||
@@ -233,7 +243,7 @@
|
||||
"reasoning": false,
|
||||
"input": ["text"],
|
||||
"contextWindow": 131072,
|
||||
"maxTokens": 16384,
|
||||
"maxTokens": 2048,
|
||||
"cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 }
|
||||
},
|
||||
{
|
||||
@@ -242,7 +252,7 @@
|
||||
"reasoning": false,
|
||||
"input": ["text", "image"],
|
||||
"contextWindow": 24576,
|
||||
"maxTokens": 8192,
|
||||
"maxTokens": 2048,
|
||||
"cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 }
|
||||
},
|
||||
{
|
||||
@@ -252,7 +262,7 @@
|
||||
"compat": { "thinkingFormat": "qwen-chat-template" },
|
||||
"input": ["text", "image"],
|
||||
"contextWindow": 24576,
|
||||
"maxTokens": 8192,
|
||||
"maxTokens": 2048,
|
||||
"cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 }
|
||||
},
|
||||
{
|
||||
@@ -262,7 +272,17 @@
|
||||
"compat": { "thinkingFormat": "qwen-chat-template" },
|
||||
"input": ["text", "image"],
|
||||
"contextWindow": 32768,
|
||||
"maxTokens": 8192,
|
||||
"maxTokens": 2048,
|
||||
"cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 }
|
||||
},
|
||||
{
|
||||
"id": "Qwen3.8-27B-UD-IQ3_XXS",
|
||||
"name": "Qwen3.8 27B · 24k · vision — hybrid reasoner",
|
||||
"reasoning": true,
|
||||
"compat": { "thinkingFormat": "qwen-chat-template" },
|
||||
"input": ["text", "image"],
|
||||
"contextWindow": 24576,
|
||||
"maxTokens": 2048,
|
||||
"cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 }
|
||||
},
|
||||
{
|
||||
@@ -272,7 +292,7 @@
|
||||
"compat": { "thinkingFormat": "qwen-chat-template" },
|
||||
"input": ["text"],
|
||||
"contextWindow": 24576,
|
||||
"maxTokens": 8192,
|
||||
"maxTokens": 2048,
|
||||
"cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 }
|
||||
},
|
||||
{
|
||||
@@ -281,7 +301,7 @@
|
||||
"reasoning": true,
|
||||
"input": ["text"],
|
||||
"contextWindow": 65536,
|
||||
"maxTokens": 8192,
|
||||
"maxTokens": 2048,
|
||||
"cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 }
|
||||
},
|
||||
{
|
||||
@@ -290,7 +310,7 @@
|
||||
"reasoning": true,
|
||||
"input": ["text"],
|
||||
"contextWindow": 65536,
|
||||
"maxTokens": 8192,
|
||||
"maxTokens": 2048,
|
||||
"cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 }
|
||||
}
|
||||
]
|
||||
|
||||
@@ -13,8 +13,8 @@
|
||||
],
|
||||
"compaction": {
|
||||
"enabled": true,
|
||||
"reserveTokens": 6144,
|
||||
"keepRecentTokens": 6000
|
||||
"reserveTokens": 2560,
|
||||
"keepRecentTokens": 1000
|
||||
},
|
||||
"retry": {
|
||||
"enabled": true,
|
||||
|
||||
Reference in New Issue
Block a user