Harald Hoyer
3a070413e4
Switch the coder model from Q6_K to the UD-Q8_K_XL quant for better output quality, and raise spec-draft-n-max from 4 to 5 to allow longer speculative draft sequences.
29 lines
732 B
INI
29 lines
732 B
INI
version = 1
|
|
|
|
[*]
|
|
flash-attn = on
|
|
parallel = 1
|
|
jinja = true
|
|
n-gpu-layers = 99
|
|
threads = 8
|
|
ubatch-size = 256
|
|
cache-type-k = bf16
|
|
cache-type-v = bf16
|
|
mmap = false
|
|
no-context-shift = true
|
|
fit = on
|
|
c = 131072
|
|
|
|
[coder]
|
|
hf = unsloth/Qwen3.6-27B-MTP-GGUF:UD-Q8_K_XL
|
|
spec-type = ngram-simple,draft-mtp
|
|
spec-draft-n-max = 5
|
|
threads-batch = 16
|
|
temp = 0.6
|
|
top-p = 0.95
|
|
top-k = 20
|
|
min-p = 0.0
|
|
parallel = 4
|
|
c = 1048576
|
|
load-on-startup = true
|
|
chat-template-kwargs = {"preserve_thinking": true}
|