harald/nixcfg
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

fix(halo): set --alias halo-8000

Browse source
This commit is contained in:
Harald Hoyer 2026-05-13 14:52:49 +02:00
parent 83b2ed3b57
commit dadfb07914
4 changed files with 4 additions and 4 deletions
Download patch file Download diff file Expand all files Collapse all files

2
systems/x86_64-linux/halo/llama-server-27B-MTP.nix
View file

@ -36,7 +36,7 @@
"--no-mmap" "--no-mmap"
"--n-gpu-layers 99" "--n-gpu-layers 99"
"-hf am17an/Qwen3.6-27B-MTP-GGUF:Q8_0" "-hf am17an/Qwen3.6-27B-MTP-GGUF:Q8_0"
"--alias qwen3.6-27b" "--alias halo-8000"
"--threads 8" "--threads 8"
"--ubatch-size 256" "--ubatch-size 256"
"-ctk bf16 -ctv bf16" "-ctk bf16 -ctv bf16"

2
systems/x86_64-linux/halo/llama-server-27B.nix
View file

@ -36,7 +36,7 @@
"--no-mmap" "--no-mmap"
"--n-gpu-layers 99" "--n-gpu-layers 99"
"-hf unsloth/Qwen3.6-27B-GGUF:UD-Q8_K_XL" "-hf unsloth/Qwen3.6-27B-GGUF:UD-Q8_K_XL"
"--alias qwen3.6-27b" "--alias halo-8000"
"--threads 8" "--threads 8"
"--ubatch-size 256" "--ubatch-size 256"
"-ctk bf16 -ctv bf16" "-ctk bf16 -ctv bf16"

2
systems/x86_64-linux/halo/llama-server-coder-next.nix
View file

@ -41,7 +41,7 @@
"--fit on" "--fit on"
"--no-context-shift" "--no-context-shift"
"-hf unsloth/Qwen3-Coder-Next-GGUF:UD-Q8_K_XL" "-hf unsloth/Qwen3-Coder-Next-GGUF:UD-Q8_K_XL"
"--alias qwen3-coder-next" "--alias halo-8000"
"--temp 1.0 --top-p 0.95 --min-p 0.01 --top-k 40" "--temp 1.0 --top-p 0.95 --min-p 0.01 --top-k 40"
"--slot-save-path %C/llama-server/kv-slots-coder-next" "--slot-save-path %C/llama-server/kv-slots-coder-next"
]; ];

2
systems/x86_64-linux/halo/llama-server.nix
View file

@ -36,7 +36,7 @@
"--no-mmap" "--no-mmap"
"--n-gpu-layers 99" "--n-gpu-layers 99"
"-hf unsloth/Qwen3.6-35B-A3B-GGUF:UD-Q8_K_XL" "-hf unsloth/Qwen3.6-35B-A3B-GGUF:UD-Q8_K_XL"
"--alias qwen3.6-35b-a3b" "--alias halo-8000"
"--threads 8" "--threads 8"
"--ubatch-size 256" "--ubatch-size 256"
"-ctk bf16 -ctv bf16" "-ctk bf16 -ctv bf16"