Lumpiasty/klaster
0
0
Fork 0
Code Issues Pull Requests 4 Packages Projects Releases Wiki Activity

fiix thinking versions of Qwen3.5 small

Browse Source
This commit is contained in:
Lumpiasty
2026-03-06 23:17:48 +01:00
parent 88a73cbb41
commit 28cb53c031
1 changed files with 17 additions and 13 deletions
Download Patch File Download Diff File Expand all files Collapse all files

30
apps/llama/configs/config.yaml
View File

@@ -566,6 +566,20 @@ models:
--no-warmup --no-warmup
--port ${PORT} --port ${PORT}
"Qwen3.5-0.8B-GGUF:Q4_K_XL":
ttl: 0
cmd: |
/app/llama-server
-hf unsloth/Qwen3.5-0.8B-GGUF:Q4_K_XL
--ctx-size 16384
--temp 0.6
--top-p 0.95
--top-k 20
--min-p 0.00
--no-warmup
--port ${PORT}
--chat-template-kwargs "{\"enable_thinking\": true}"
"Qwen3.5-0.8B-GGUF-nothink:Q4_K_XL": "Qwen3.5-0.8B-GGUF-nothink:Q4_K_XL":
ttl: 0 ttl: 0
cmd: | cmd: |
@@ -580,19 +594,6 @@ models:
--port ${PORT} --port ${PORT}
--chat-template-kwargs "{\"enable_thinking\": false}" --chat-template-kwargs "{\"enable_thinking\": false}"
"Qwen3.5-0.8B-GGUF:Q4_K_XL":
ttl: 0
cmd: |
/app/llama-server
-hf unsloth/Qwen3.5-0.8B-GGUF:Q4_K_XL
--ctx-size 16384
--temp 0.6
--top-p 0.95
--top-k 20
--min-p 0.00
--no-warmup
--port ${PORT}
"Qwen3.5-2B-GGUF:Q4_K_M": "Qwen3.5-2B-GGUF:Q4_K_M":
ttl: 600 ttl: 600
cmd: | cmd: |
@@ -605,6 +606,7 @@ models:
--min-p 0.00 --min-p 0.00
--no-warmup --no-warmup
--port ${PORT} --port ${PORT}
--chat-template-kwargs "{\"enable_thinking\": true}"
"Qwen3.5-2B-GGUF-nothink:Q4_K_M": "Qwen3.5-2B-GGUF-nothink:Q4_K_M":
ttl: 600 ttl: 600
@@ -632,6 +634,7 @@ models:
--min-p 0.00 --min-p 0.00
--no-warmup --no-warmup
--port ${PORT} --port ${PORT}
--chat-template-kwargs "{\"enable_thinking\": true}"
"Qwen3.5-4B-GGUF-nothink:Q4_K_M": "Qwen3.5-4B-GGUF-nothink:Q4_K_M":
ttl: 600 ttl: 600
@@ -659,6 +662,7 @@ models:
--min-p 0.00 --min-p 0.00
--no-warmup --no-warmup
--port ${PORT} --port ${PORT}
--chat-template-kwargs "{\"enable_thinking\": true}"
"Qwen3.5-9B-GGUF-nothink:Q4_K_M": "Qwen3.5-9B-GGUF-nothink:Q4_K_M":
ttl: 600 ttl: 600