; llama-server --host 0.0.0.0 --port 8082 --models-max 1 --models-preset backend.ini --ui-config-file frontend.json

[*]
no-mmap = on               ; Disable mmap (default: off)
mlock = on                 ; Lock model in RAM
fit = off                  ; Disable automatic memory fitting (default: on)
ngl = 999                  ; Full GPU offload
ctk = q8_0                 ; KV cache key quantization
ctv = q8_0                 ; KV cache value quantization
fa = on                    ; Enable flash attention
np = 4                     ; Parallel request batching (default: 4)
kvu = on                   ; Unified KV cache buffer (default: on)
stop-timeout = 2           ; Force-kill child process after graceful shutdown timeout in seconds (default: 10)
sleep-idle-seconds = 300   ; Unload weights on child process
b = 4096                   ; Logical maximum batch size (default: 2048)
ub = 2048                  ; Physical maximum batch size (default: 512)
webui-mcp-proxy = on       ; Enabled for llama-server issue debugging, not exposed to the internet
; slot-save-path = /var/www/ia/kvcache

; The last user message always gets an anchor, min-step only gates the intermediate ones (no anchor right after an mtmd chunk)
ctx-checkpoints = 32       ; Recurrent/SWA state anchors per slot, oldest evicted first (default: 32)
checkpoint-min-step = 4096 ; Minimum token spacing between anchors, 32 x 4096 = 131072 token span (default: 8192)
cache-ram = 16384          ; 16 GiB for idle slot states, anchors included (default: 8192)

[MoE-Qwen3.8-Flash-Next-125B-A6B]
m = unsloth/Qwen3.8-Flash-Next-GGUF/Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00004.gguf
c = 262144
ot = per_layer_token_embd=CPU

[MoE-GPT-OSS-120B]
m = lmstudio-community/gpt-oss-120b-GGUF/gpt-oss-120b-MXFP4-00001-of-00002.gguf
; load-on-startup = 1     ; Load immediately on server startup
c = 131072              ; Context size in tokens for this model
; alias = gpt-4, gpt-4o

[MoE-GPT-OSS-20B]
m = lmstudio-community/gpt-oss-20b-GGUF/gpt-oss-20b-MXFP4.gguf
c = 131072

[MoE-Laguna-S-2.1-118B-A8B]
m = unsloth/Laguna-S-2.1-GGUF/Laguna-S-2.1-UD-Q4_K_XL-00001-of-00003.gguf
c = 262144

[MoE-Ling-Flash-2.0-100B]
m = bartowski/inclusionAI_Ling-flash-2.0-GGUF/inclusionAI_Ling-flash-2.0-Q6_K-00001-of-00003.gguf
rope-scaling = yarn
rope-scale = 4.0
yarn-orig-ctx = 32768
override-kv = bailingmoe2.context_length=int:131072
c = 131072

[MoE-Solar-Open-100B]
m = mradermacher/Solar-Open-100B-i1-GGUF/Solar-Open-100B.i1-Q6_K.gguf
c = 131072

[Dense-Vision-Qwen3.8-27B]
m = unsloth/Qwen3.8-27B-GGUF/Qwen3.8-27B-UD-Q8_K_XL.gguf
mm = unsloth/Qwen3.8-27B-GGUF/mmproj-BF16.gguf
reasoning-effort = medium
temp = 1.0
top-p = 0.95
top-k = 20
min-p = 0
presence-penalty = 0
repeat-penalty = 1
c = 262144

[Dense-Vision-Qwen3.8-27B-MTP]
m = unsloth/Qwen3.8-27B-GGUF/Qwen3.8-27B-UD-Q8_K_XL.gguf
mm = unsloth/Qwen3.8-27B-GGUF/mmproj-BF16.gguf
reasoning-effort = medium
spec-type = draft-mtp
spec-draft-n-max = 3
temp = 1.0
top-p = 0.95
top-k = 20
min-p = 0
presence-penalty = 0
repeat-penalty = 1
c = 262144

[Dense-Uncensored-Qwen3.8-27B]
m = mradermacher/Qwen3.8-27B-heretic-ara-GGUF/Qwen3.8-27B-heretic-ara.Q8_0.gguf
mm = mradermacher/Qwen3.8-27B-heretic-ara-GGUF/Qwen3.8-27B-heretic-ara.mmproj-f16.gguf
reasoning-effort = medium
temp = 1.0
top-p = 0.95
top-k = 20
min-p = 0
presence-penalty = 0
repeat-penalty = 1
c = 262144

[Dense-Vision-Qwen3.6-27B]
m = unsloth/Qwen3.6-27B-GGUF/Qwen3.6-27B-UD-Q8_K_XL.gguf
mm = unsloth/Qwen3.6-27B-GGUF/mmproj-BF16.gguf
temp = 0.6
top-p = 0.96
top-k = 20
min-p = 0
presence-penalty = 0
repeat-penalty = 1
c = 262144

[Dense-Vision-Qwen3.6-27B-MTP]
m = unsloth/Qwen3.6-27B-MTP-GGUF/Qwen3.6-27B-UD-Q8_K_XL.gguf
mm = unsloth/Qwen3.6-27B-MTP-GGUF/mmproj-BF16.gguf
spec-type = draft-mtp
spec-draft-n-max = 3
temp = 0.6
top-p = 0.96
top-k = 20
min-p = 0
presence-penalty = 0
repeat-penalty = 1
c = 262144

[MoE-Qwen3-30B-A3B-Instruct-2507]
m = unsloth/Qwen3-30B-A3B-Instruct-2507-GGUF/Qwen3-30B-A3B-Instruct-2507-UD-Q8_K_XL.gguf
temp = 0.7
top-p = 0.8
top-k = 20
min-p = 0
c = 262144

[MoE-Qwen3-30B-A3B-Thinking-2507]
m = unsloth/Qwen3-30B-A3B-Thinking-2507-GGUF/Qwen3-30B-A3B-Thinking-2507-UD-Q8_K_XL.gguf
temp = 0.6
top-p = 0.95
top-k = 20
min-p = 0
c = 262144

[MoE-Vision-Qwen3.6-35B-A3B]
m = unsloth/Qwen3.6-35B-A3B-GGUF/Qwen3.6-35B-A3B-UD-Q8_K_XL.gguf
mm = unsloth/Qwen3.6-35B-A3B-GGUF/mmproj-BF16.gguf
temp = 0.6
top-p = 0.96
top-k = 20
min-p = 0
presence-penalty = 0
repeat-penalty = 1
c = 262144

; [MoE-Vision-Qwen3.6-35B-A3B-MTP]
; m = unsloth/Qwen3.6-35B-A3B-MTP-GGUF/Qwen3.6-35B-A3B-UD-Q8_K_XL.gguf
; mm = unsloth/Qwen3.6-35B-A3B-MTP-GGUF/mmproj-BF16.gguf
; spec-type = draft-mtp
; spec-draft-n-max = 3
; temp = 0.6
; top-p = 0.96
; top-k = 20
; min-p = 0
; presence-penalty = 0
; repeat-penalty = 1
; c = 262144

[MoE-Vision-Qwen3.5-122B-A10B]
m = unsloth/Qwen3.5-122B-A10B-GGUF/Qwen3.5-122B-A10B-UD-Q4_K_XL-00001-of-00003.gguf
mm = unsloth/Qwen3.5-122B-A10B-GGUF/mmproj-BF16.gguf
c = 262144

[MoE-Vision-Mistral-Small-4-119B-2603]
m = unsloth/Mistral-Small-4-119B-2603-GGUF/Mistral-Small-4-119B-2603-UD-Q4_K_XL-00001-of-00003.gguf
mm = unsloth/Mistral-Small-4-119B-2603-GGUF/mmproj-BF16.gguf
c = 262144

[Dense-Vision-Mistral-Medium-3.5-128B]
m = unsloth/Mistral-Medium-3.5-128B-GGUF/Mistral-Medium-3.5-128B-UD-Q4_K_XL-00001-of-00003.gguf
mm = unsloth/Mistral-Medium-3.5-128B-GGUF/mmproj-BF16.gguf
; c = 131072
c = 16384

[Dense-Uncensored-Dolphin-Mistral-24B-Venice-Edition]
m = bartowski/cognitivecomputations_Dolphin-Mistral-24B-Venice-Edition-GGUF/cognitivecomputations_Dolphin-Mistral-24B-Venice-Edition-Q8_0.gguf
c = 65536

[Dense-Uncensored-BlackSheep-24B]
m = mradermacher/BlackSheep-24B-GGUF/BlackSheep-24B.Q8_0.gguf
c = 65536

[Dense-Vision-Gemma-4-31B-IT]
m = unsloth/gemma-4-31B-it-GGUF/gemma-4-31B-it-UD-Q8_K_XL.gguf
mm = unsloth/gemma-4-31B-it-GGUF/mmproj-BF16.gguf
c = 262144

[Dense-Vision-Gemma-4-31B-IT-MTP]
m = unsloth/gemma-4-31B-it-GGUF/gemma-4-31B-it-UD-Q8_K_XL.gguf
mm = unsloth/gemma-4-31B-it-GGUF/mmproj-BF16.gguf
md = unsloth/gemma-4-31B-it-GGUF/gemma-4-31B-it-MTP-BF16.gguf
spec-type = draft-mtp
spec-draft-n-max = 4
c = 262144

[MoE-Vision-Gemma-4-26B-A4B-IT]
m = unsloth/gemma-4-26B-A4B-it-GGUF/gemma-4-26B-A4B-it-UD-Q8_K_XL.gguf
mm = unsloth/gemma-4-26B-A4B-it-GGUF/mmproj-BF16.gguf
c = 262144

; [MoE-Vision-Gemma-4-26B-A4B-IT-MTP]
; m = unsloth/gemma-4-26B-A4B-it-GGUF/gemma-4-26B-A4B-it-UD-Q8_K_XL.gguf
; mm = unsloth/gemma-4-26B-A4B-it-GGUF/mmproj-BF16.gguf
; md = unsloth/gemma-4-26B-A4B-it-GGUF/gemma-4-26B-A4B-it-MTP-BF16.gguf
; spec-type = draft-mtp
; spec-draft-n-max = 2
; c = 262144

[Dense-Vision-Gemma-4-12b-IT]
m = unsloth/gemma-4-12b-it-GGUF/gemma-4-12b-it-UD-Q8_K_XL.gguf
mm = unsloth/gemma-4-12b-it-GGUF/mmproj-BF16.gguf
c = 262144

[Dense-Vision-Gemma-4-12b-IT-MTP]
m = unsloth/gemma-4-12b-it-GGUF/gemma-4-12b-it-UD-Q8_K_XL.gguf
mm = unsloth/gemma-4-12b-it-GGUF/mmproj-BF16.gguf
md = unsloth/gemma-4-12b-it-GGUF/gemma-4-12b-it-BF16-MTP.gguf
spec-type = draft-mtp
spec-draft-n-max = 2
c = 262144

[MoE-Vision-Gemma-4-E2B-IT]
m = unsloth/gemma-4-E2B-it-GGUF/gemma-4-E2B-it-UD-Q8_K_XL.gguf
mm = unsloth/gemma-4-E2B-it-GGUF/mmproj-BF16.gguf
c = 262144

[MoE-Vision-Gemma-4-E2B-IT-MTP]
m = unsloth/gemma-4-E2B-it-GGUF/gemma-4-E2B-it-UD-Q8_K_XL.gguf
mm = unsloth/gemma-4-E2B-it-GGUF/mmproj-BF16.gguf
md = unsloth/gemma-4-E2B-it-GGUF/gemma-4-E2B-it-BF16-MTP.gguf
spec-type = draft-mtp
spec-draft-n-max = 2
c = 262144

[MoE-Vision-Gemma-4-E4B-IT]
m = unsloth/gemma-4-E4B-it-GGUF/gemma-4-E4B-it-UD-Q8_K_XL.gguf
mm = unsloth/gemma-4-E4B-it-GGUF/mmproj-BF16.gguf
c = 262144

[MoE-Vision-Gemma-4-E4B-IT-MTP]
m = unsloth/gemma-4-E4B-it-GGUF/gemma-4-E4B-it-UD-Q8_K_XL.gguf
mm = unsloth/gemma-4-E4B-it-GGUF/mmproj-BF16.gguf
md = unsloth/gemma-4-E4B-it-GGUF/gemma-4-E4B-it-BF16-MTP.gguf
spec-type = draft-mtp
spec-draft-n-max = 2
c = 262144

[MoE-GLM-4.5-Air-106B]
m = unsloth/GLM-4.5-Air-GGUF/GLM-4.5-Air-UD-Q5_K_XL-00001-of-00002.gguf
c = 131072

[MoE-INTELLECT-3-106B]
m = bartowski/PrimeIntellect_INTELLECT-3-GGUF/PrimeIntellect_INTELLECT-3-Q5_K_M-00001-of-00003.gguf
c = 131072

[MoE-Vision-GLM-4.6V-106B]
m = unsloth/GLM-4.6V-GGUF/GLM-4.6V-UD-Q5_K_XL-00001-of-00002.gguf
mm = unsloth/GLM-4.6V-GGUF/mmproj-BF16.gguf
c = 131072

[MoE-Vision-Step-3.7-Flash-197B-A11B]
m = unsloth/Step-3.7-Flash-GGUF/Step-3.7-Flash-UD-Q2_K_XL-00001-of-00003.gguf
mm = unsloth/Step-3.7-Flash-GGUF/mmproj-BF16.gguf
c = 262144

[MoE-GLM-4.7-Flash-30B-A3B]
m = unsloth/GLM-4.7-Flash-GGUF/GLM-4.7-Flash-UD-Q8_K_XL.gguf
temp = 0.7
top-p = 1.0
min-p = 0.01
c = 202752

[MoE-NVIDIA-Nemotron-3.5-Lightning-30B-A3B]
m = unsloth/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-GGUF/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-UD-Q8_K_XL.gguf
temp = 1.0
top-p = 0.95
c = 262144

[Dense-Llama-3.3-70B-Instruct]
m = unsloth/Llama-3.3-70B-Instruct-GGUF/Llama-3.3-70B-Instruct-UD-Q6_K_XL-00001-of-00002.gguf
c = 131072

[Dense-Llama-3_3-Nemotron-Super-49B-v1_5]
m = unsloth/Llama-3_3-Nemotron-Super-49B-v1_5-GGUF/Llama-3_3-Nemotron-Super-49B-v1_5-UD-Q8_K_XL-00001-of-00002.gguf
c = 131072

[MoE-NVIDIA-Nemotron-3-Super-120B-A12B]
m = unsloth/NVIDIA-Nemotron-3-Super-120B-A12B-GGUF/NVIDIA-Nemotron-3-Super-120B-A12B-UD-Q4_K_XL-00001-of-00003.gguf
c = 262144

[MoE-MiniMax-M2.7-229B]
m = unsloth/MiniMax-M2.7-GGUF/MiniMax-M2.7-UD-Q2_K_XL-00001-of-00003.gguf
c = 131072

[MoE-MiniMax-M3-426B]
mlock = off
m = unsloth/MiniMax-M3-GGUF/MiniMax-M3-UD-Q2_K_XL-00001-of-00004.gguf
n-cpu-moe = 28
c = 131072

[MoE-DeepSeek-V4-Flash-0731-284B-A13B-Q8_K_XL-POC]
m = unsloth/DeepSeek-V4-Flash-0731-GGUF/DeepSeek-V4-Flash-0731-UD-Q8_K_XL-00001-of-00005.gguf
load-mode = none       ; No mmap, no mlock: streaming manages its own residency
moe-stream-cache = 65  ; VRAM expert budget, GiB: pinned partition + dynamic wave slots
moe-stream-direct = on ; O_DIRECT reads, page cache bypassed
c = 262144

[MoE-DeepSeek-V4-Flash-0731-284B-A13B-Q8_K_XL]
mlock = off
m = unsloth/DeepSeek-V4-Flash-0731-GGUF/DeepSeek-V4-Flash-0731-UD-Q8_K_XL-00001-of-00005.gguf
n-cpu-moe = 20
c = 262144

; [MoE-DeepSeek-V4-Flash-0731-284B-A13B-Q8_K_XL-DSpark]
; mlock = off
; m = unsloth/DeepSeek-V4-Flash-0731-GGUF/DeepSeek-V4-Flash-0731-UD-Q8_K_XL-00001-of-00005.gguf
; md = unsloth/DeepSeek-V4-Flash-0731-GGUF/dspark-DeepSeek-V4-Flash-0731-Q8_0.gguf
; spec-type = draft-dspark
; spec-draft-n-max = 5
; n-cpu-moe = 24
; c = 262144

[MoE-DeepSeek-V4-Flash-0731-284B-A13B-IQ2_XXS]
m = unsloth/DeepSeek-V4-Flash-0731-GGUF/DeepSeek-V4-Flash-0731-UD-IQ2_XXS-00001-of-00003.gguf
c = 262144

[Dense-Granite-4.0-H-Small-32B]
m = unsloth/granite-4.0-h-small-GGUF/granite-4.0-h-small-UD-Q8_K_XL.gguf
c = 131072

[MoE-North-Mini-Code-1.0-30B-A3B]
m = unsloth/North-Mini-Code-1.0-GGUF/North-Mini-Code-1.0-UD-Q8_K_XL.gguf
c = 262144

[Dense-Muse-Glimmer-30B]
m = unsloth/Muse-Glimmer-30B-GGUF/Muse-Glimmer-30B-UD-Q8_K_XL.gguf
mm = unsloth/Muse-Glimmer-30B-GGUF/mmproj-Muse-Glimmer-30B-BF16.gguf
temp = 1.0
top-p = 0.95
top-k = 64
min-p = 0
repeat-penalty = 1
presence-penalty = 0
c = 131072

[Dense-Muse-Glimmer-30B-DFlash]
m = unsloth/Muse-Glimmer-30B-GGUF/Muse-Glimmer-30B-UD-Q8_K_XL.gguf
mm = unsloth/Muse-Glimmer-30B-GGUF/mmproj-Muse-Glimmer-30B-BF16.gguf
md = unsloth/Muse-Glimmer-30B-GGUF/dflash-kquant.gguf
spec-type = draft-dflash
spec-draft-n-max = 15
temp = 1.0
top-p = 0.95
top-k = 64
min-p = 0
repeat-penalty = 1
presence-penalty = 0
c = 131072