LM/Qwen 3.8 27B
< LM
Jump to navigation
Jump to search
llama.cpp
Running
# works
./llama.cpp/sycl/bin/llama-server \
-hf unsloth/Qwen3.8-27B-GGUF:UD-Q6_K_M \
--spec-type draft-mtp --spec-draft-n-max 2 -ngld 99 \
-c 64000 -ctk q8_0 -ctv q8_0 \
-fa on -ngl 99 --load-mode mlock -np 1 --jinja --reasoning-preserve -t 8
# with turboquant
./llama.cpp/vulkan/bin/llama-server \
-hf unsloth/Qwen3.8-27B-GGUF:UD-Q6_K_M \
--spec-type draft-mtp --spec-draft-n-max 2 -ngld 99 \
-c 64000 -ctk q8_0 -ctv turbo4 \
-fa on -ngl 99 --load-mode mlock -np 1 --jinja --reasoning-preserve -t 8
Benchmark for me
# control group
./llama.cpp/sycl/bin/llama-bench \
-hf unsloth/Qwen3.8-27B-GGUF:UD-Q5_K_XL \
-fa on -ctk q8_0 -ctv q8_0 -ngl 99
| model | size | params | backend | ngl | type_k | type_v | fa | test | t/s |
| ------------------------------ | ---------: | ---------: | ---------- | --: | -----: | -----: | --: | --------------: | -------------------: |
| qwen35 27B Q5_K - Medium | 19.43 GiB | 27.32 B | SYCL | 99 | q8_0 | q8_0 | 1 | pp512 | 951.05 ± 7.72 |
| qwen35 27B Q5_K - Medium | 19.43 GiB | 27.32 B | SYCL | 99 | q8_0 | q8_0 | 1 | tg128 | 22.17 ± 0.06 |
# Test -ctv
./llama.cpp/sycl/bin/llama-bench \
-hf unsloth/Qwen3.8-27B-GGUF:UD-Q5_K_XL \
-fa on -ctk q8_0 -ctv q4_0 -ngl 99
# Test -ub
./llama.cpp/sycl/bin/llama-bench \
-hf unsloth/Qwen3.8-27B-GGUF:UD-Q5_K_XL \
-fa on -ctk q8_0 -ctv q8_0 -ngl 99 -ub 2048
# Test -fa off
./llama.cpp/sycl/bin/llama-bench \
-hf unsloth/Qwen3.8-27B-GGUF:UD-Q5_K_XL \
-fa off -ngl 99
# Test bartowski
./llama.cpp/sycl/bin/llama-bench \
-hf bartowski/Qwen3.8-27B-GGUF:Q5_K_L \
-fa on -ctk q8_0 -ctv q8_0 -ngl 99
| model | size | params | backend | ngl | type_k | type_v | fa | test | t/s |
| ------------------------------ | ---------: | ---------: | ---------- | --: | -----: | -----: | --: | --------------: | -------------------: |
| qwen35 27B Q5_K - Medium | 20.05 GiB | 27.32 B | SYCL | 99 | q8_0 | q8_0 | 1 | pp512 | 997.64 ± 2.90 |
| qwen35 27B Q5_K - Medium | 20.05 GiB | 27.32 B | SYCL | 99 | q8_0 | q8_0 | 1 | tg128 | 22.08 ± 0.06 |
# Test OpenVINO
GGML_OPENVINO_DEVICE=GPU GGML_OPENVINO_STATEFUL_EXECUTION=1 ./llama.cpp/openvino/bin/llama-bench \
-hf unsloth/Qwen3.8-27B-GGUF:Q4_0 \
-fa on -ngl 99
Benchmark for community
./llama.cpp/sycl/bin/llama-bench -hf bartowski/Qwen3.8-27B-GGUF:IQ3_XXS -fa 0,1
| model | size | params | backend | threads | fa | test | t/s |
| ------------------------------ | ---------: | ---------: | ---------- | ------: | --: | --------------: | -------------------: |
| qwen35 27B IQ3_XXS - 3.0625 bpw | 11.75 GiB | 27.32 B | SYCL,BLAS | 6 | 0 | pp512 | 445.32 ± 0.53 |
| qwen35 27B IQ3_XXS - 3.0625 bpw | 11.75 GiB | 27.32 B | SYCL,BLAS | 6 | 0 | tg128 | 10.28 ± 0.01 |
| qwen35 27B IQ3_XXS - 3.0625 bpw | 11.75 GiB | 27.32 B | SYCL,BLAS | 6 | 1 | pp512 | 436.31 ± 0.21 |
| qwen35 27B IQ3_XXS - 3.0625 bpw | 11.75 GiB | 27.32 B | SYCL,BLAS | 6 | 1 | tg128 | 10.23 ± 0.03 |
build: d7a207411 (10644)
./llama.cpp/sycl/bin/llama-bench -hf bartowski/Qwen3.8-27B-GGUF:Q4_0 -fa 0,1
| model | size | params | backend | threads | fa | test | t/s |
| ------------------------------ | ---------: | ---------: | ---------- | ------: | --: | --------------: | -------------------: |
| qwen35 27B Q4_0 | 15.22 GiB | 27.32 B | SYCL,BLAS | 6 | 0 | pp512 | 812.57 ± 2.08 |
| qwen35 27B Q4_0 | 15.22 GiB | 27.32 B | SYCL,BLAS | 6 | 0 | tg128 | 26.89 ± 0.03 |
| qwen35 27B Q4_0 | 15.22 GiB | 27.32 B | SYCL,BLAS | 6 | 1 | pp512 | 354.67 ± 1.02 |
| qwen35 27B Q4_0 | 15.22 GiB | 27.32 B | SYCL,BLAS | 6 | 1 | tg128 | 26.77 ± 0.02 |
build: d7a207411 (10644)
./llama.cpp/sycl/bin/llama-bench -hf bartowski/Qwen3.8-27B-GGUF:Q4_K_L -fa 0,1
| model | size | params | backend | threads | fa | test | t/s |
| ------------------------------ | ---------: | ---------: | ---------- | ------: | --: | --------------: | -------------------: |
| qwen35 27B Q4_K - Medium | 17.42 GiB | 27.32 B | SYCL,BLAS | 6 | 0 | pp512 | 867.21 ± 1.71 |
| qwen35 27B Q4_K - Medium | 17.42 GiB | 27.32 B | SYCL,BLAS | 6 | 0 | tg128 | 24.97 ± 0.09 |
| qwen35 27B Q4_K - Medium | 17.42 GiB | 27.32 B | SYCL,BLAS | 6 | 1 | pp512 | 756.78 ± 38.06 |
| qwen35 27B Q4_K - Medium | 17.42 GiB | 27.32 B | SYCL,BLAS | 6 | 1 | tg128 | 24.86 ± 0.09 |
build: d7a207411 (10644)
./llama.cpp/sycl/bin/llama-bench -hf bartowski/Qwen3.8-27B-GGUF:Q5_K_L -fa 0,1
| model | size | params | backend | threads | fa | test | t/s |
| ------------------------------ | ---------: | ---------: | ---------- | ------: | --: | --------------: | -------------------: |
| qwen35 27B Q5_K - Medium | 20.05 GiB | 27.32 B | SYCL,BLAS | 6 | 0 | pp512 | 1031.14 ± 4.23 |
| qwen35 27B Q5_K - Medium | 20.05 GiB | 27.32 B | SYCL,BLAS | 6 | 0 | tg128 | 22.24 ± 0.03 |
| qwen35 27B Q5_K - Medium | 20.05 GiB | 27.32 B | SYCL,BLAS | 6 | 1 | pp512 | 833.60 ± 0.72 |
| qwen35 27B Q5_K - Medium | 20.05 GiB | 27.32 B | SYCL,BLAS | 6 | 1 | tg128 | 22.11 ± 0.03 |
build: d7a207411 (10644)
./llama.cpp/sycl/bin/llama-bench -hf bartowski/Qwen3.8-27B-GGUF:Q6_K_L -fa 0,1
| model | size | params | backend | threads | fa | test | t/s |
| ------------------------------ | ---------: | ---------: | ---------- | ------: | --: | --------------: | -------------------: |
| qwen35 27B Q6_K | 22.42 GiB | 27.32 B | SYCL,BLAS | 6 | 0 | pp512 | 984.56 ± 5.09 |
| qwen35 27B Q6_K | 22.42 GiB | 27.32 B | SYCL,BLAS | 6 | 0 | tg128 | 20.08 ± 0.03 |
| qwen35 27B Q6_K | 22.42 GiB | 27.32 B | SYCL,BLAS | 6 | 1 | pp512 | 715.19 ± 1.40 |
| qwen35 27B Q6_K | 22.42 GiB | 27.32 B | SYCL,BLAS | 6 | 1 | tg128 | 19.93 ± 0.09 |
build: d7a207411 (10644)
./llama.cpp/sycl/bin/llama-bench -hf unsloth/Qwen3.8-27B-GGUF:Q4_0 -fa 0,1
| model | size | params | backend | threads | fa | test | t/s |
| ------------------------------ | ---------: | ---------: | ---------- | ------: | --: | --------------: | -------------------: |
| qwen35 27B Q4_0 | 14.94 GiB | 27.32 B | SYCL,BLAS | 6 | 0 | pp512 | 815.63 ± 1.66 |
| qwen35 27B Q4_0 | 14.94 GiB | 27.32 B | SYCL,BLAS | 6 | 0 | tg128 | 27.59 ± 0.03 |
| qwen35 27B Q4_0 | 14.94 GiB | 27.32 B | SYCL,BLAS | 6 | 1 | pp512 | 358.96 ± 1.03 |
| qwen35 27B Q4_0 | 14.94 GiB | 27.32 B | SYCL,BLAS | 6 | 1 | tg128 | 27.40 ± 0.03 |
build: d7a207411 (10644)
./llama.cpp/sycl/bin/llama-bench -hf unsloth/Qwen3.8-27B-GGUF:UD-Q4_K_XL -fa 0,1
| model | size | params | backend | threads | fa | test | t/s |
| ------------------------------ | ---------: | ---------: | ---------- | ------: | --: | --------------: | -------------------: |
| qwen35 27B Q4_K - Medium | 16.34 GiB | 27.32 B | SYCL,BLAS | 6 | 0 | pp512 | 727.84 ± 20.79 |
| qwen35 27B Q4_K - Medium | 16.34 GiB | 27.32 B | SYCL,BLAS | 6 | 0 | tg128 | 21.38 ± 0.05 |
| qwen35 27B Q4_K - Medium | 16.34 GiB | 27.32 B | SYCL,BLAS | 6 | 1 | pp512 | 754.42 ± 6.23 |
| qwen35 27B Q4_K - Medium | 16.34 GiB | 27.32 B | SYCL,BLAS | 6 | 1 | tg128 | 21.32 ± 0.05 |
build: d7a207411 (10644)
./llama.cpp/sycl/bin/llama-bench -hf unsloth/Qwen3.8-27B-GGUF:UD-Q5_K_XL -fa 0,1
| model | size | params | backend | threads | fa | test | t/s |
| ------------------------------ | ---------: | ---------: | ---------- | ------: | --: | --------------: | -------------------: |
| qwen35 27B Q5_K - Medium | 19.43 GiB | 27.32 B | SYCL,BLAS | 6 | 0 | pp512 | 978.10 ± 6.44 |
| qwen35 27B Q5_K - Medium | 19.43 GiB | 27.32 B | SYCL,BLAS | 6 | 0 | tg128 | 22.31 ± 0.03 |
| qwen35 27B Q5_K - Medium | 19.43 GiB | 27.32 B | SYCL,BLAS | 6 | 1 | pp512 | 843.67 ± 2.55 |
| qwen35 27B Q5_K - Medium | 19.43 GiB | 27.32 B | SYCL,BLAS | 6 | 1 | tg128 | 22.24 ± 0.02 |
build: d7a207411 (10644)
./llama.cpp/sycl/bin/llama-bench -hf unsloth/Qwen3.8-27B-GGUF:UD-Q6_K -fa 0,1
| model | size | params | backend | threads | fa | test | t/s |
| ------------------------------ | ---------: | ---------: | ---------- | ------: | --: | --------------: | -------------------: |
| qwen35 27B Q6_K | 20.46 GiB | 27.32 B | SYCL,BLAS | 6 | 0 | pp512 | 973.08 ± 5.22 |
| qwen35 27B Q6_K | 20.46 GiB | 27.32 B | SYCL,BLAS | 6 | 0 | tg128 | 21.21 ± 0.02 |
| qwen35 27B Q6_K | 20.46 GiB | 27.32 B | SYCL,BLAS | 6 | 1 | pp512 | 789.89 ± 2.14 |
| qwen35 27B Q6_K | 20.46 GiB | 27.32 B | SYCL,BLAS | 6 | 1 | tg128 | 21.15 ± 0.02 |
build: d7a207411 (10644)
Experiment
# search
hf models ls --search "Qwen3.8-27B" --apps llama.cpp --expand "downloads,likes,lastModified" --sort downloads --no-truncate --limit 10
# unsloth
hf models ls --tree -R -h unsloth/Qwen3.8-27B-GGUF
hf download unsloth/Qwen3.8-27B-GGUF --include "*UD-IQ4_XS*"
hf download unsloth/Qwen3.8-27B-GGUF --include "*UD-Q4_K_XL*"
hf download unsloth/Qwen3.8-27B-GGUF --include "*UD-Q5_K_XL*"
hf download unsloth/Qwen3.8-27B-GGUF --include "*UD-Q6_K.gguf"
hf download unsloth/Qwen3.8-27B-GGUF --include "*UD-Q6_K_M*"
hf download unsloth/Qwen3.8-27B-GGUF --include "*UD-Q6_K_L*"
hf download unsloth/Qwen3.8-27B-GGUF --include "*UD-Q6_K_XL*"
hf download unsloth/Qwen3.8-27B-GGUF --include "*UD-Q8_K_L*"
hf download unsloth/Qwen3.8-27B-GGUF --include "*Q4_0*"
hf download unsloth/Qwen3.8-27B-GGUF --include "*Q8_0*"
hf download unsloth/Qwen3.8-27B-GGUF --include "mmproj-BF16*"
hf download unsloth/Qwen3.8-27B-GGUF --include "*mtp-Qwen3.8-27B-Q4_0.gguf"
find ~/.cache/huggingface/hub -name "*Qwen3.8-27B-UD-Q8_K_L.gguf"
find ~/.cache/huggingface/hub -name "*Qwen3.8-27B-UD-Q8_K_L.gguf" -exec sh -c 'rm -f "$1" "$(readlink -f "$1")"' _ {} \;
tree -h ~/.cache/huggingface/hub/models--unsloth--Qwen3.8-27B-GGUF
# bartowski
hf models ls --tree -R -h bartowski/Qwen3.8-27B-GGUF
hf download bartowski/Qwen3.8-27B-GGUF --include "*IQ3_XXS*"
hf download bartowski/Qwen3.8-27B-GGUF --include "*IQ4_XS*"
hf download bartowski/Qwen3.8-27B-GGUF --include "*Q4_0*"
hf download bartowski/Qwen3.8-27B-GGUF --include "*Q4_1*"
hf download bartowski/Qwen3.8-27B-GGUF --include "*Q4_K_L*"
hf download bartowski/Qwen3.8-27B-GGUF --include "*Q5_K_L*"
hf download bartowski/Qwen3.8-27B-GGUF --include "*Q6_K_L*"
hf download bartowski/Qwen3.8-27B-GGUF --include "*Q8_0*"
hf download bartowski/Qwen3.8-27B-GGUF --include "*mmproj-Qwen3.8-27B-bf16.gguf*"
tree -h ~/.cache/huggingface/hub/models--bartowski--Qwen3.8-27B-GGUF
