[LLM] models update
This commit is contained in:
@@ -150,6 +150,8 @@
|
||||
"gemma-4-26B-A4B-it-UD-IQ4_XS": { "name": "Gemma 4 26B · 24k · vision — quality generalist (~34 t/s, 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 (~57 t/s, remote)", "attachment": true, "limit": { "context": 65536, "output": 16384 } },
|
||||
"GLM-4.7-Flash-UD-Q4_K_XL": { "reasoning": true, "interleaved": "reasoning_content", "name": "GLM-4.7 Flash · 24k — quality coder (~21 t/s, remote)", "limit": { "context": 24576, "output": 8192 } },
|
||||
"GLM-4.7-Flash-UD-Q3_K_XL": { "reasoning": true, "interleaved": "reasoning_content", "name": "GLM-4.7 Flash Q3 · 24k — fast coder (~28 t/s, remote)", "limit": { "context": 24576, "output": 8192 } },
|
||||
"Ornith-1.5-35B-A3B-IQ3_XXS": { "reasoning": true, "name": "Ornith 1.5 35B · 24k · vision — agentic coder, thinking (~32 t/s, remote)", "attachment": true, "limit": { "context": 24576, "output": 8192 } },
|
||||
"gpt-oss-20b": { "reasoning": true, "name": "gpt-oss 20B · 64k — fast reasoning + tools (~38 t/s, remote)", "limit": { "context": 65536, "output": 8192 } },
|
||||
"gpt-oss-20b-low": { "reasoning": true, "name": "gpt-oss 20B low · 64k — snappy answers (~37 t/s, remote)", "limit": { "context": 65536, "output": 8192 } }
|
||||
}
|
||||
@@ -171,6 +173,8 @@
|
||||
"gemma-4-26B-A4B-it-UD-IQ4_XS": { "name": "Gemma 4 26B · 24k · vision — quality generalist (~34 t/s)", "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 (~57 t/s)", "attachment": true, "limit": { "context": 65536, "output": 16384 } },
|
||||
"GLM-4.7-Flash-UD-Q4_K_XL": { "reasoning": true, "interleaved": "reasoning_content", "name": "GLM-4.7 Flash · 24k — quality coder (~21 t/s)", "limit": { "context": 24576, "output": 8192 } },
|
||||
"GLM-4.7-Flash-UD-Q3_K_XL": { "reasoning": true, "interleaved": "reasoning_content", "name": "GLM-4.7 Flash Q3 · 24k — fast coder (~28 t/s)", "limit": { "context": 24576, "output": 8192 } },
|
||||
"Ornith-1.5-35B-A3B-IQ3_XXS": { "reasoning": true, "name": "Ornith 1.5 35B · 24k · vision — agentic coder, thinking (~32 t/s)", "attachment": true, "limit": { "context": 24576, "output": 8192 } },
|
||||
"gpt-oss-20b": { "reasoning": true, "name": "gpt-oss 20B · 64k — fast reasoning + tools (~38 t/s)", "limit": { "context": 65536, "output": 8192 } },
|
||||
"gpt-oss-20b-low": { "reasoning": true, "name": "gpt-oss 20B low · 64k — snappy answers (~37 t/s)", "limit": { "context": 65536, "output": 8192 } }
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user