library_name: llama.cpp
pipeline_tag: text-generation
tags:
- gguf
- nvfp4
- mtp
- qwen
Qwen3.6-27B-Uncensored-HauhauCS-Aggressive-NVFP4-MTP-GGUF
Files
| File | Source lineage | Size |
|---|---|---|
Qwen3.6-27B-Uncensored-HauhauCS-Aggressive-NVFP4-MTP-Q4_K_P.gguf |
MTP Q4_K_P source GGUF |
15,388,070,304 bytes |
Qwen3.6-27B-Uncensored-HauhauCS-Aggressive-NVFP4-MTP-Q6_K_P.gguf |
MTP Q6_K_P source GGUF |
15,388,070,368 bytes |
Qwen3.6-27B-Uncensored-HauhauCS-Aggressive-NVFP4-MTP-Q8_K_P.gguf |
MTP Q8_K_P source GGUF |
15,388,070,368 bytes |
Q4_K_P, Q6_K_P, and Q8_K_P describe the source GGUF lineage. The files themselves are NVFP4 GGUF files.
Conversion Summary
All variants preserve the original MTP layer as block 64 and keep F32 tensors as F32.
Q4_K_P-derived artifact
325 Q4_K tensors -> NVFP4
173 Q6_K tensors -> NVFP4
8 Q8_0 tensors -> NVFP4
360 F32 tensors -> F32
Q6_K_P-derived artifact
374 Q6_K tensors -> NVFP4
132 Q8_0 tensors -> NVFP4
360 F32 tensors -> F32
Q8_K_P-derived artifact
407 Q8_0 tensors -> NVFP4
99 F16 tensors -> NVFP4
360 F32 tensors -> F32
Validation
Each MTP artifact was checked after conversion:
tensor_count = 866
NVFP4 = 506
F32 = 360
qwen35.block_count = 65
qwen35.nextn_predict_layers = 1
mtp_tensors = 15
general.file_type = 39
Optimized Agentic Tooling
Tool-calling on llama.cpp with CUDA 13.3+ using WebUI.
Q8_K_P
llama-server.exe ^
-m "Qwen3.6-27B-Uncensored-HauhauCS-Aggressive-NVFP4-MTP-Q8_K_P.gguf" ^
--jinja ^
--spec-type draft-mtp ^
--spec-draft-n-max 1 ^
--spec-draft-ngl 100 ^
-ngl 100 ^
-np 1 ^
-fa on ^
-c 262144 ^
-ctk q4_0 ^
-ctv q4_0 ^
--context-shift ^
--host 127.0.0.1 ^
--port 8033 ^
--tools read_file,file_glob_search,grep_search,exec_shell_command,write_file,edit_file,apply_diff
Q6_K_P
llama-server.exe ^
-m "Qwen3.6-27B-Uncensored-HauhauCS-Aggressive-NVFP4-MTP-Q6_K_P.gguf" ^
--jinja ^
--spec-type draft-mtp ^
--spec-draft-n-max 1 ^
--spec-draft-ngl 100 ^
-ngl 100 ^
-np 1 ^
-fa on ^
-c 262144 ^
-ctk q4_0 ^
-ctv q4_0 ^
--context-shift ^
--host 127.0.0.1 ^
--port 8033 ^
--tools read_file,file_glob_search,grep_search,exec_shell_command,write_file,edit_file,apply_diff
Q4_K_P
llama-server.exe ^
-m "Qwen3.6-27B-Uncensored-HauhauCS-Aggressive-NVFP4-MTP-Q4_K_P.gguf" ^
--jinja ^
--spec-type draft-mtp ^
--spec-draft-n-max 2 ^
--spec-draft-ngl 100 ^
-ngl 100 ^
-np 1 ^
-fa on ^
-c 262144 ^
-ctk q4_0 ^
-ctv q4_0 ^
--context-shift ^
--host 127.0.0.1 ^
--port 8033 ^
--tools read_file,file_glob_search,grep_search,exec_shell_command,write_file,edit_file,apply_diff