turintech/Qwen3.5-0.8B-Artemis-GGUF

TurinTech models, built with Artemis

A tiny Qwen3.5 0.8B for local tool calls and short chat answers, with all layers on a small GPU.

• 0.569 GB download for the default file, Qwen3.5-0.8B-Artemis-Q5_K_M.gguf.
• Peak GPU memory 1.2 GiB on an NVIDIA RTX 5090 with an 8,192-token context; the original peaked at 2.2 GiB.
• 163 of 258 held-out BFCL live tool calls right, against 109 for bartowski's Q4_K_M (0.580 GB).

Run it

Needs llama.cpp with Qwen3.5 and HTTPS support. -hf also downloads the vision projector and loads it; add --no-mmproj for text only, which skips the 207 MB download.

llama-server -hf turintech/Qwen3.5-0.8B-Artemis-GGUF:Q5_K_M --port 8080 -c 8192 --jinja

Request example, images, settings and Ollama

Keep the :Q5_K_M tag: untagged llama.cpp downloads and Hugging Face's generated Ollama import select Q4_K_M instead.

llama-server -m Qwen3.5-0.8B-Artemis-Q5_K_M.gguf --port 8080 -c 8192 --jinja

-c 8192 sets an 8,192-token context; --jinja uses the chat template stored in the file, which is Qwen's original. The raw template defaults to thinking off when enable_thinking is absent. llama-server can supply its own default; in our b11617 runtime check, an omitted flag enabled thinking. In our test, a short question with thinking on used all 1,500 output tokens thinking and returned no answer, so turn thinking off in the request for short answers and tool calls, as below.

curl http://localhost:8080/v1/chat/completions -H 'Content-Type: application/json' -d '{
  "messages": [{"role": "user", "content": "What is the weather in Paris right now, in celsius?"}],
  "tools": [{"type": "function", "function": {"name": "get_weather", "description": "Current weather for a city",
    "parameters": {"type": "object", "properties": {"city": {"type": "string"},
      "unit": {"type": "string", "enum": ["celsius", "fahrenheit"]}}, "required": ["city"]}}}],
  "chat_template_kwargs": {"enable_thinking": false}
}'

In our test run the reply's tool_calls held get_weather with {"city":"Paris","unit":"celsius"}.

# the smaller file: download it, then the same command
hf download turintech/Qwen3.5-0.8B-Artemis-GGUF Qwen3.5-0.8B-Artemis-Q4_K_M.gguf --local-dir .
llama-server -m Qwen3.5-0.8B-Artemis-Q4_K_M.gguf --port 8080 -c 8192 --jinja

Questions, problems and requests for other small models are welcome in our feedback thread.

Images

The vision projector is a separate file, converted from the original weights. mmproj-F16.gguf is the same projector in F16.

hf download turintech/Qwen3.5-0.8B-Artemis-GGUF mmproj-BF16.gguf --local-dir .
llama-server -m Qwen3.5-0.8B-Artemis-Q5_K_M.gguf --mmproj mmproj-BF16.gguf --port 8080 -c 8192 --jinja

Send the image as a data URL in an image_url part of the user message. On 20 chart questions this file with the BF16 projector got 11 right (11 with the F16 one), against 11 for Unsloth's Q5_K_M with Unsloth's projector (see Image questions).

Ollama

Read-only registry checks on 16 October 2026 confirmed the tag aliases, model-layer SHA-256 hashes and package settings below against these HF files. This was not a fresh registry pull or inference test.

ollama run turintech/qwen3.5-0.8b-artemis:q5_k_m

latest and q5_k_m are the Q5_K_M file and q4_k_m is the Q4_K_M file, with the same SHA-256 as this repository. The packages use Ollama's Qwen3.5 renderer and tool-call parser, as Ollama's own qwen3.5 package does, and Qwen's recommended sampling for text in thinking mode (temperature 1, top_k 20, top_p 0.95, presence penalty 1.5). To import a file directly with the same settings:

hf download turintech/Qwen3.5-0.8B-Artemis-GGUF Qwen3.5-0.8B-Artemis-Q5_K_M.gguf --local-dir .
printf 'FROM ./Qwen3.5-0.8B-Artemis-Q5_K_M.gguf\nRENDERER qwen3.5\nPARSER qwen3.5\nPARAMETER temperature 1\nPARAMETER top_k 20\nPARAMETER top_p 0.95\nPARAMETER presence_penalty 1.5\n' > Modelfile
ollama create qwen3.5-0.8b-artemis -f Modelfile
ollama run qwen3.5-0.8b-artemis

A separate Ollama 0.40.4 local-import smoke test covered our audited files with one tool prompt, greedy decoding, seed 20261015 and a 256-token output cap. In this test, omitting think enabled reasoning; think:false disabled it. This checks import and parser behavior on one prompt, not task accuracy or a fresh registry pull. The accuracy comparisons use llama.cpp with thinking explicitly off.

The packages hold the main file only. For images, use Ollama 0.40.11 or newer and pull from Hugging Face, which brings the repository's projector along:

ollama run hf.co/turintech/Qwen3.5-0.8B-Artemis-GGUF:Q5_K_M

With Ollama 0.40.11 on 26 October, the same pull from Unsloth's repository of this model brought its projector along and ollama show listed vision. A local package of our Q5_K_M file and mmproj-F16.gguf also listed vision, answered 11 of the 20 chart questions and returned get_weather with Paris and celsius for the tool request above. To build that package, put both files next to a Modelfile with two FROM lines:

FROM ./Qwen3.5-0.8B-Artemis-Q5_K_M.gguf
FROM ./mmproj-F16.gguf
RENDERER qwen3.5
PARSER qwen3.5
PARAMETER num_ctx 8192

Choose this model

• Best for tool calls and short chat answers on a small GPU.
• Consider another file when knowledge questions matter: on MMLU-Pro it got 33 of 280 right against 44 for the original, and Unsloth's Q8_0 (0.812 GB) stays far closer to the original.

More on choosing

On the RTX 5090 GSM8K sample (problems 0 to 249, a guardrail in choosing the Q4_K_M), Q5_K_M answered 132/250 correctly and Q4_K_M 137/250, versus 145/250 for BF16. On IFEval (all 541 prompts, strict), which played no part in choosing the files, Q5_K_M followed 308 and Q4_K_M 296, versus 290 for BF16; no paired test separated either file from BF16. Over all 400 tool-calling tasks, every one of which was used in choosing the files, Q5_K_M got 279 right against 292 for BF16 and 290 and 282 for the public Q5_K_M files, while Q4_K_M got 313.

The results do not establish overall superiority or lossless compression. See the measured trade-offs and test conditions in How it was tested.

Files

Every file we publish for this model. Sizes are download sizes, not memory.

File	Size	Use it for	Ollama tag
Qwen3.5-0.8B-Artemis-Q5_K_M.gguf	569 MB	Default	q5_k_m (latest)
Qwen3.5-0.8B-Artemis-Q4_K_M.gguf	532 MB	Smaller download	q4_k_m
Qwen3.5-0.8B-Artemis-imatrix.gguf	2 MB	Importance matrix used to make the files; for reference, not a model to run	n/a
mmproj-BF16.gguf	207 MB	Vision companion (mmproj): pass it with --mmproj for image inputs	n/a
mmproj-F16.gguf	205 MB	Vision companion (mmproj): pass it with --mmproj for image inputs	n/a

• With llama.cpp or a Hugging Face pull, add the file's quant label as the tag: -hf turintech/Qwen3.5-0.8B-Artemis-GGUF:Q5_K_M.
• The Ollama tag runs our package, which holds the same file: ollama run turintech/qwen3.5-0.8b-artemis:<tag>.
• LM Studio's Hub entry turintech/qwen3.5-0.8b-artemis offers every model file above, without the importance matrix or the vision projector.

Without a tag, -hf and hf.co fetch Qwen3.5-0.8B-Artemis-Q4_K_M.gguf, not the default file, so always add the tag.

The vision projector is optional, for image input. There is no MTP file.

How the files compare

File	Size	Choose for
Qwen3.5-0.8B-Artemis-Q5_K_M.gguf	0.569 GB	Default.
Qwen3.5-0.8B-Artemis-Q4_K_M.gguf	0.532 GB	Smaller download.

GB means decimal download size, not RAM or VRAM. Labels are size classes; tensor types and checksums are below.

Results

Our default file against bartowski's Q4_K_M, the closest public file in size, and the original. These tasks played no part in choosing the files.

File	Size	Tool calls (of 258)	IFEval (of 541)	GSM8K (of 250)	MMLU-Pro (of 280)
Ours, Q5_K_M	0.569 GB	163	308	124	33
bartowski Q4_K_M	0.580 GB	109	285	114	39
BF16 original	1.52 GB	153	290	120	44

Only the tool-call gap over bartowski's file survives correction. MMLU-Pro is the weak spot: 33 of 280 against 44 for the original. How it was tested.

Correct tool calls on the 258 BFCL live_simple tasks, an independent panel that no recorded selection run used, measured on an NVIDIA RTX 5090 with llama.cpp b11617 and drawn on the full 0 to 258 scale. Our Q4_K_M got the most right, 173 against 153 for the original and 115 for Unsloth's Q4_K_M, and our Q5_K_M 163 against 109 for bartowski's Q4_K_M; these three differences survive correction. The files with fewer right made no tool call more often. No other pre-registered test on the panel separated our files from the original or the matched public files after correction; among those, the largest gap against ours was MMLU-Pro, where the Q5_K_M got 33 of 280 right against 44 for the original. Every task expects a call, so the chart does not show whether a file holds back when no tool fits. See How it was tested.

Memory

The upstream architecture declares a 262,144-token context limit; this audit does not validate these GGUFs at that maximum. A separate Q4_K_M retrieval probe recovered three codes from repetitive filler at 12,067 and 24,067 input tokens, using a 32,768-token server context and seed 2026101504. This is not long-document reasoning evidence. YaRN extension and quantized KV caches were not evaluated. The benchmark comparisons used F16 K and V caches.

Peak memory was measured on an NVIDIA RTX 5090 with llama.cpp b11617, all layers on the GPU, an 8,192-token context, F16 K and V caches and one slot. Each file ran in its own llama-server process with the prompt cache off (--cache-ram 0) and answered the first 25 items of each benchmark task. GPU memory is the peak from nvidia-smi; host memory is the server's peak resident memory, which counts the parts of the memory-mapped model file that were read.

File	Size	Peak GPU memory	Peak host memory
BF16 original	1.52 GB	2.2 GiB	1.9 GiB
Artemis Q5_K_M	0.569 GB	1.2 GiB	1.0 GiB
Artemis Q4_K_M	0.532 GB	1.3 GiB	0.9 GiB
Unsloth Q8_0	0.812 GB	1.5 GiB	1.2 GiB
bartowski Q5_K_M	0.646 GB	1.4 GiB	1.0 GiB
Unsloth Q5_K_M	0.590 GB	1.3 GiB	1.0 GiB
bartowski Q4_K_M	0.580 GB	1.3 GiB	1.0 GiB
Unsloth UD-Q4_K_XL	0.559 GB	1.3 GiB	1.0 GiB
Unsloth Q4_K_M	0.533 GB	1.2 GiB	0.9 GiB
bartowski IQ4_XS	0.523 GB	1.3 GiB	1.0 GiB

A smaller file does not always need less memory: our Q4_K_M peaked higher on the GPU than our Q5_K_M. The commands on this card keep llama-server's default 8 GiB prompt cache, which lives in host memory: over the full benchmark, host memory grew to 9.0 to 9.6 GiB for every file as the cache filled, while GPU memory stayed the same. Add --cache-ram 0 to the command to stay near the table's host figures. CPU-only memory was not measured. The context length and other runtime settings change these numbers.

With the projector loaded (mmproj-F16.gguf, RTX 5090, llama.cpp b11889, -ngl 99 -c 8192 -np 1 --cache-ram 0), 1.7 GiB of GPU memory was in use after the 20 chart questions.

How it was tested

Measurements, comparisons and limitations

Every file, ours and public, was compared with a BF16 conversion of the same original (Qwen/Qwen3.5-0.8B revision 2fc0636, converted without the MTP layer):

• KL divergence and next-token agreement over 60 chunks of 512 tokens from each of two human-written texts: the WikiText-2 test set (Wikipedia) and OpenAssistant OASST1 validation conversations.
• Native tool calling on BFCL simple (Python): 400 tasks, each a user request and one function to call. The first 200 were used to choose between candidate files and the other 200 were reserved initially from that search. Each request went through llama-server's OpenAI-style tools API with one slot, greedy decoding, thinking off and Qwen's original chat template for every file, and was scored with BFCL's own checker. A reply with no tool call counts as wrong.
• GSM8K maths, 250 problems (the first 250 of the test set on the RTX 5090).
• MMLU-Pro, 280 questions (20 from each category), zero-shot with a direct letter answer and replies capped at 8 tokens. This is not the leaderboard protocol. A reply without a letter counts as wrong, and even the original gave no letter on 117 of 280, so these counts mix answer format with knowledge.
• IFEval, all 541 prompts, strict prompt-level accuracy with the lm-evaluation-harness v0.4.13 checks.

The two KL texts were used in choosing both files, so the KL and same-next-token columns are not independent tests. MMLU-Pro and IFEval played no part in choosing the files. GSM8K problems 0 to 249 served as a pass/fail guardrail in the search that produced the Q4_K_M; GSM8K played no part in choosing the Q5_K_M. In October 2026 the RTX 5090 suite was rerun on a second RTX 5090 machine with the same protocol and llama.cpp build. The rerun used its own implementation of that protocol. It reproduced every earlier value in the RTX 5090 tables at the precision shown, and added the count over all 400 tool-calling tasks, MMLU-Pro, IFEval and five more public files.

The selection-exposed 200 are not fresh. On the Apple M3 Max they helped decide which files to publish. On the RTX 5090 they were run on four final candidates for the Q4_K_M file, and the file here is the one with the most tool calls right among them, so its count may be higher than it would be on new tasks. The October rerun reproduced every selection-panel count.

The paired test compares two files on the same tasks. It counts the tasks only one of the two got right and gives the chance of a split at least that uneven if the files were equally good (exact McNemar test). A large p means the test found no difference, which is not proof that the files are equal.

NVIDIA RTX 5090, llama.cpp b11617 (CUDA)

File	Size	KL wiki / chat	Same next token, wiki / chat	Tool calls right (of 400)	Tool calls right, first 200	Tool calls right, selection-exposed 200	No tool call at all
BF16 original	1.52 GB	reference	reference	292	136	156	22
Artemis Q5_K_M	0.569 GB	.0098 / .0092	94.2% / 95.1%	279	119	160	13
Artemis Q4_K_M	0.532 GB	.0165 / .0163	92.7% / 93.6%	313	149	164	12
Unsloth Q8_0	0.812 GB	.0012 / .0012	98.1% / 98.2%	307	149	158	20
bartowski Q5_K_M	0.646 GB	.0101 / .0096	94.4% / 95.0%	282	132	150	29
Unsloth Q5_K_M	0.590 GB	.0120 / .0108	94.0% / 94.8%	290	140	150	20
bartowski Q4_K_M	0.580 GB	.0297 / .0295	90.8% / 91.9%	173	74	99	89
Unsloth UD-Q4_K_XL	0.559 GB	.0237 / .0238	91.8% / 92.8%	226	104	122	55
Unsloth Q4_K_M	0.533 GB	.0338 / .0333	90.1% / 91.4%	162	70	92	95
bartowski IQ4_XS	0.523 GB	.0487 / .0474	88.7% / 89.3%	88	42	46	147

"No tool call at all" counts replies on the selection-exposed 200 that made no tool call.

File	GSM8K (of 250)	MMLU-Pro (of 280)	IFEval strict (of 541)
BF16 original	145	38	290
Artemis Q5_K_M	132	33	308
Artemis Q4_K_M	137	37	296
Unsloth Q8_0	147	34	295
bartowski Q5_K_M	136	29	288
Unsloth Q5_K_M	137	29	295
bartowski Q4_K_M	138	31	285
Unsloth UD-Q4_K_XL	131	25	291
Unsloth Q4_K_M	131	33	276
bartowski IQ4_XS	126	8	294

Comparison on the 200-task selection panel	Only the first right	Only the second right	p
Artemis Q5_K_M vs bartowski Q4_K_M	64	3	7e-16
Artemis Q4_K_M vs Unsloth Q4_K_M	73	1	8e-21
Artemis Q5_K_M vs BF16 original	9	5	0.42 (not significant)
Artemis Q4_K_M vs BF16 original	11	3	0.06 (not significant)
Artemis Q4_K_M vs Artemis Q5_K_M	8	4	0.39 (not significant)

Comparisons with the original and with files added in October, on each half of the 400 tool-calling tasks and on all of them. Each cell gives the tasks only the first file got right, the tasks only the second got right, and p, uncorrected.

Comparison (exact McNemar)	Tasks 0-199	Tasks 200-399	All 400
Artemis Q5_K_M vs BF16 original	1 / 18, p<0.001	9 / 5, p=0.42	10 / 23, p=0.035
Artemis Q4_K_M vs BF16 original	21 / 8, p=0.024	11 / 3, p=0.057	32 / 11, p=0.002
Artemis Q5_K_M vs Unsloth Q5_K_M	4 / 25, p<0.001	13 / 3, p=0.021	17 / 28, p=0.14
Artemis Q5_K_M vs bartowski Q5_K_M	4 / 17, p=0.007	15 / 5, p=0.041	19 / 22, p=0.76
Artemis Q5_K_M vs Unsloth Q8_0	1 / 31, p<0.001	7 / 5, p=0.77	8 / 36, p<0.001
Artemis Q4_K_M vs Unsloth UD-Q4_K_XL	50 / 5, p<0.001	44 / 2, p<0.001	94 / 7, p<0.001

The same paired test on GSM8K, MMLU-Pro and IFEval. MMLU-Pro and IFEval played no part in choosing the files; GSM8K served as the guardrail described above for the Q4_K_M. Each cell gives the items only the first file got right, the items only the second got right, and p.

Comparison (exact McNemar)	GSM8K (250)	MMLU-Pro (280)	IFEval strict (541)
Artemis Q5_K_M vs BF16 original	20 / 33, p=0.10	4 / 9, p=0.27	55 / 37, p=0.08
Artemis Q4_K_M vs BF16 original	24 / 32, p=0.35	8 / 9, p=1.00	48 / 42, p=0.60
Artemis Q5_K_M vs Unsloth Q5_K_M	21 / 26, p=0.56	10 / 6, p=0.45	55 / 42, p=0.22
Artemis Q5_K_M vs bartowski Q5_K_M	21 / 25, p=0.66	8 / 4, p=0.39	56 / 36, p=0.047
Artemis Q5_K_M vs Unsloth Q8_0	17 / 32, p=0.044	4 / 5, p=1.00	48 / 35, p=0.19
Artemis Q4_K_M vs Unsloth Q4_K_M	37 / 31, p=0.54	13 / 9, p=0.52	73 / 53, p=0.09
Artemis Q4_K_M vs bartowski Q4_K_M	32 / 33, p=1.00	13 / 7, p=0.26	70 / 59, p=0.38

The October run made 64 paired tests over the four tasks, with tool calls counted on all 400 tasks, every one of which was used in choosing our files. After a Bonferroni correction for those 64, the differences that remain are on tool calls, plus MMLU-Pro against bartowski's IQ4_XS, which gave no letter on 254 of its 280 replies. On the 400 tool-calling tasks, our two files got more right than the plain Q4_K_M, UD-Q4_K_XL and IQ4_XS files, our Q4_K_M also got more right than bartowski's Q5_K_M, and Unsloth's Q8_0 got more right than our Q5_K_M (36 against 8, mostly in the first 200). No GSM8K or IFEval difference survives the correction, so the card does not claim an IFEval gain. The selection-panel rows and the per-half columns above are outside those 64 tests; every p in those tables is shown uncorrected.

The table reports selection-panel correct-call counts for Q4_K_M and the original. The paired test does not separate the two (p = 0.06), so the card does not claim that it beats the original. Tool-calling counts also move a lot between task sets: on the first 200 tasks the Q5_K_M file got 119 right against 136 for the original, with 58 replies that made no tool call, while on the selection-exposed 200 the paired test did not separate it from the original. The first 200 were the search set for the Q4_K_M file, so its 149 there is not a fair comparison either. The original RTX 5090 rows came from one harness on GPUs of the same model; re-run on a second GPU, the original and the Q4_K_M file scored the same on every selection-exposed task, and the October rerun reproduced all of those rows.

Independent panel, RTX 5090, October 2026

A later run on the machine used for the October rerun, with the same protocol and llama.cpp build (14310190c, b11617), measured the ten files above on tasks that no recorded selection run used. They are independent as far as the records go; unrecorded earlier use cannot be ruled out. It used the same prompts, reply caps and checkers as the tasks above:

• GSM8K fresh: the last 250 test problems (rows 1069 to 1318). The searches behind our files used problems 0 to 255 at most.
• MMLU-Pro fresh: 280 other questions (20 per category, seed 20261022), none of them in the sample above.
• BFCL live_simple: all 258 tasks, user-contributed requests with one function to call. No selection record for these files uses this category; 60 of the tasks ran in the fresh CPU probe below, after the files were chosen.
• IFEval: the 541 prompts above, from the October rerun.

Before the run, six comparisons were fixed for this card: each of our files against the original, Q4_K_M against Unsloth's Q4_K_M, and Q5_K_M against bartowski's Q4_K_M, Unsloth's Q5_K_M and bartowski's Q5_K_M, each on the four tasks. Only these 24 tests, after a Bonferroni correction for 24, can support a stated difference between two files.

File	Size	GSM8K fresh (of 250)	MMLU-Pro fresh (of 280)	BFCL live_simple (of 258)	IFEval strict (of 541)	No tool call at all
BF16 original	1.52 GB	120	44	153	290	40
Artemis Q5_K_M	0.569 GB	124	33	163	308	17
Artemis Q4_K_M	0.532 GB	123	38	173	296	11
Unsloth Q8_0	0.812 GB	128	45	150	295	38
bartowski Q5_K_M	0.646 GB	124	27	151	288	40
Unsloth Q5_K_M	0.590 GB	124	38	161	295	27
bartowski Q4_K_M	0.580 GB	114	39	109	285	109
Unsloth UD-Q4_K_XL	0.559 GB	125	33	145	291	46
Unsloth Q4_K_M	0.533 GB	116	35	115	276	95
bartowski IQ4_XS	0.523 GB	116	12	67	294	170

Here "No tool call at all" counts replies to the 258 live tasks that made no tool call. The 768-token cap cut off 45 of the original's 250 GSM8K fresh replies, against 31 for our Q5_K_M and 32 for our Q4_K_M.

Comparison (exact McNemar)	GSM8K fresh (250)	MMLU-Pro fresh (280)	BFCL live_simple (258)	IFEval strict (541)
Artemis Q5_K_M vs BF16 original	29 / 25, p=0.68	2 / 13, p=0.007	20 / 10, p=0.099	55 / 37, p=0.08
Artemis Q4_K_M vs BF16 original	34 / 31, p=0.80	5 / 11, p=0.21	25 / 5, p<0.001	48 / 42, p=0.60
Artemis Q4_K_M vs Unsloth Q4_K_M	41 / 34, p=0.49	13 / 10, p=0.68	65 / 7, p<0.001	73 / 53, p=0.09
Artemis Q5_K_M vs bartowski Q4_K_M	39 / 29, p=0.27	11 / 17, p=0.34	63 / 9, p<0.001	72 / 49, p=0.045
Artemis Q5_K_M vs Unsloth Q5_K_M	28 / 28, p=1.00	10 / 15, p=0.42	15 / 13, p=0.85	55 / 42, p=0.22
Artemis Q5_K_M vs bartowski Q5_K_M	31 / 31, p=1.00	13 / 7, p=0.26	21 / 9, p=0.043	56 / 36, p=0.047

Each cell gives the items only the first file got right, the items only the second got right, and p, uncorrected. After the Bonferroni correction for these 24 tests, three differences remain, all on live tool calls: our Q4_K_M against the original (25 / 5, corrected p=0.008) and against Unsloth's Q4_K_M (65 / 7, corrected p<0.001), and our Q5_K_M against bartowski's Q4_K_M (63 / 9, corrected p<0.001). The files with fewer right made no tool call more often: 40 times for the original, 95 for Unsloth's Q4_K_M and 109 for bartowski's, against 11 and 17 for ours. Every live_simple task expects a call, so the panel does not test whether a file holds back when no tool fits. No other test survives. The Q5_K_M answered fewer MMLU-Pro fresh questions than the original (2 / 13, 0.18 corrected), and at this size many MMLU-Pro replies gave no letter: 113 of 280 for the original, 137 for our Q5_K_M and 114 for our Q4_K_M.

Apple M3 Max, llama.cpp b11617 (Metal)

File	Size	KL wiki / chat	Tool calls right (of 200)	GSM8K (of 250)	MMLU-Pro (of 280)
BF16 original	1.52 GB	reference	142	134	39
Artemis Q5_K_M (Mac build)	0.569 GB	.0087 / .0086	140	122	32
bartowski Q4_K_M	0.580 GB	.0274 / .0279	65	129	33
Unsloth Q4_K_M	0.533 GB	.0309 / .0291	76	124	37

Comparison on the 200-task selection panel	Only the first right	Only the second right	p
Artemis Q5_K_M (Mac build) vs bartowski Q4_K_M	76	1	1e-21
Artemis Q5_K_M (Mac build) vs BF16 original	12	14	0.85 (not significant)

The Q4_K_M file in this repository was chosen after the M3 Max run and has not been run on that machine.

Compare rows within one table. The two machines used separate harnesses for the same tests, so the prompts and the chat text sample may not match exactly. The M3 Max rows for our Q5_K_M are a separate build made on the Mac, with the same tensor types and its own importance matrix. It is not the Q5_K_M file published here, which was built on the RTX 5090 machine. The two machines give different absolute counts: the BF16 original got 142 selection-exposed tasks right on the M3 Max and 156 on the RTX 5090, and every file's KL is 6 to 14% higher on the RTX 5090. Both machines' tables include our Q5_K_M, bartowski's Q4_K_M and the original, with paired tests between them. On the selection-exposed 200 neither machine separated Q5_K_M from the original; on the RTX 5090's first 200 the original got 18 tasks right that Q5_K_M missed, against 1. The RTX 5090 runs scored tool calls with the answer key in bfcl_eval 2026.3.23 and the M3 Max run with the BFCL v3 key; the two keys differ on one selection-exposed task.

Most of the tool-calling losses in the public 4-bit files are replies with no tool call at all. On the RTX 5090 selection-exposed 200 these were 89 for bartowski's Q4_K_M, 95 for Unsloth's Q4_K_M, 55 for UD-Q4_K_XL and 147 for IQ4_XS, against 22 for the original.

Fresh CPU probe, 15 October 2026

This separate check used workstation CPU, eight inference threads, llama.cpp b11617, one 8,192-token server slot, greedy decoding (seed 20261015), thinking explicitly off and the pinned public Unsloth BF16 reference. KL is the mean of six document means: two recent Wikipedia extracts, two authored code samples and two authored chats, each evaluated as two 256-token chunks. These short, fixed probes measure next-token fidelity, including the reference's mistakes.

Tool scores use 60 BFCL live_simple cases selected with seed 2026101502, from a category that no selection record for these files uses; unrecorded earlier use cannot be ruled out. The table reports exact two-sided McNemar p-values. Download sizes use decimal GB; size differences are explicit.

Comparison (ours / public GB)	Mean KL ours / public	Lower mean KL	Correct calls /60	Only ours / public	McNemar p
Our Q4_K_M / Unsloth Q4_K_M (0.532 / 0.533)	0.022265 / 0.039005	42.9%	39 / 28	13 / 2	0.007385
Our Q5_K_M / bartowski Q4_K_M (0.569 / 0.580)	0.013993 / 0.034045	58.9%	39 / 28	14 / 3	0.01273
Our Q5_K_M / Unsloth UD-Q4_K_XL (0.569 / 0.559)	0.013993 / 0.025893	46.0%	39 / 39	7 / 7	1
Our Q5_K_M / Unsloth Q5_K_S (0.569 / 0.569)	0.013993 / 0.018341	23.7%	39 / 43	4 / 8	0.3877

BF16 scored 34/60. In table order, six-document sign-flip p-values are 0.03125, 0.03125, 0.03125, 0.03125. BFCL clustered p-values are 0.019287, 0.011353, 1, 0.34375. The clusters are the 41 function groups defined by BFCL task IDs. After Bonferroni correction across 16 comparisons per endpoint, the KL p-values are 0.5, 0.5, 0.5, 0.5. None of these tool-call differences survives that correction. These are exploratory fixed-file comparisons, not evidence of general superiority or equivalence. The fourth comparator is a post-hoc coverage check added after inspecting download sizes, before inspecting its outputs.

Lower mean KL does not mean every token is closer: on prose-literature, our Q4 maximum token KL is 2.481113 versus Unsloth Q4 1.248548. The advantage over plain Q4 files does not extend to Unsloth’s dynamic Q4 on this tool slice: both scored 39/60. The similarly sized Unsloth Q5_K_S scored 43/60 against our Q5 at 39/60; the lower mean KL of our Q5 did not translate into a higher tool score.

Corrected authored tool diagnostic

A second tool set contains 24 authored prompts in six families, with explicit date/time, enum and exact-copy constraints. Its schemas were corrected after inspecting ambiguous v1 outputs, so this is not a blind benchmark. The rerun used greedy decoding, thinking off, a 1,024-token cap and seed 2026101506.

File	Correct /24
Our Q4_K_M	24
Our Q5_K_M	23
Unsloth BF16	23
Unsloth Q4_K_M	20
Unsloth UD-Q4_K_XL	22
Unsloth Q5_K_S	24
bartowski Q4_K_M	18

In the four comparison rows above, exact six-family sign-flip p-values are 0.25, 0.125, 1, 1, and Bonferroni-adjusted values across 16 diagnostic comparisons are 1, 1, 1, 1. A high score on these few related prompts does not establish general tool accuracy or equivalence.

Separate CPU replay of the original panels

The released files were rerun on the workstation CPU with llama.cpp b11617, eight inference threads, one 8,192-token slot, greedy decoding, seed 20261015 and thinking off. BFCL uses the original 400 simple-Python tasks and official AST checker, with the development and selection-exposed halves kept separate. GSM8K uses the exact original first 250 questions, a 768-token output limit and the original boxed-answer/last-number parser. The public Unsloth BF16 file is the reference; the historical producer BF16 identity is unestablished. These results do not recreate the separate unpublished Mac Q5 build or establish CPU/GPU equivalence.

File	BFCL development /200	BFCL selection-exposed /200	GSM8K /250	Boxed-only /250	Nontruncated /250
Our Q4_K_M	148	161	136	135	135
Our Q5_K_M	119	153	138	136	137
unsloth BF16	136	156	143	142	142
unsloth Q4_K_M	75	94	133	130	130
unsloth UD-Q4_K_XL	104	119	not run	not run	not run
bartowski Q4_K_M	74	97	137	136	136

Boxed-only and nontruncated sensitivities keep all 250 questions in the denominator and treat rejected outputs as failures. They were specified after inspecting truncation and parser behavior. The historical RTX BF16 result is 145/250; this CPU/public-reference result is 143/250, with 35 output-limit hits. Counts can change across artifact, hardware and runtime settings.

GSM8K comparison	Only ours / other	Exact McNemar p	Bonferroni p (10 pairs)
Our Q4_K_M / unsloth BF16	26 / 33	0.43499	1
Our Q4_K_M / unsloth Q4_K_M	32 / 29	0.79815	1
Our Q4_K_M / bartowski Q4_K_M	35 / 36	1	1
Our Q5_K_M / unsloth BF16	23 / 28	0.57585	1
Our Q5_K_M / unsloth Q4_K_M	33 / 28	0.60892	1
Our Q5_K_M / bartowski Q4_K_M	33 / 32	1	1

These post-hoc paired tests concern this fixed panel and scorer, not general maths quality.

CPU BFCL paired tests

Comparison	Panel	Only ours / other	Exact McNemar p	Bonferroni p (30 tests)
Our Q4_K_M / unsloth BF16	dev200	19 / 7	0.028959	0.86878
Our Q4_K_M / unsloth BF16	selection-exposed200	11 / 6	0.33231	1
Our Q4_K_M / unsloth Q4_K_M	dev200	76 / 3	2.7211e-19	8.1633e-18
Our Q4_K_M / unsloth Q4_K_M	selection-exposed200	70 / 3	1.3743e-17	4.1228e-16
Our Q4_K_M / unsloth UD-Q4_K_XL	dev200	48 / 4	1.3065e-10	3.9196e-09
Our Q4_K_M / unsloth UD-Q4_K_XL	selection-exposed200	45 / 3	1.3126e-10	3.9378e-09
Our Q4_K_M / bartowski Q4_K_M	dev200	77 / 3	1.4128e-19	4.2385e-18
Our Q4_K_M / bartowski Q4_K_M	selection-exposed200	66 / 2	1.5904e-17	4.7712e-16
Our Q5_K_M / unsloth BF16	dev200	4 / 21	0.00091052	0.027316
Our Q5_K_M / unsloth BF16	selection-exposed200	9 / 12	0.66362	1
Our Q5_K_M / unsloth Q4_K_M	dev200	47 / 3	3.7083e-11	1.1125e-09
Our Q5_K_M / unsloth Q4_K_M	selection-exposed200	62 / 3	2.4842e-15	7.4527e-14
Our Q5_K_M / unsloth UD-Q4_K_XL	dev200	23 / 8	0.010674	0.32022
Our Q5_K_M / unsloth UD-Q4_K_XL	selection-exposed200	37 / 3	1.9465e-08	5.8395e-07
Our Q5_K_M / bartowski Q4_K_M	dev200	47 / 2	4.3556e-12	1.3067e-10
Our Q5_K_M / bartowski Q4_K_M	selection-exposed200	59 / 3	1.7249e-14	5.1748e-13

The correction covers all 15 file pairs on both panels. Neither panel is new independent release-selection evidence.

The original WikiText/OASST texts were independently reconstructed to matching hashes and replayed over 60 chunks of 512 tokens each. These are new CPU measurements with the pinned public BF16 reference, separately from the historical hardware/build tables above.

File	Wiki / chat mean KL	Wiki / chat top-1 agreement	Wiki / chat P99 token KL
bartowski Q4_K_M	0.029953 / 0.029502	90.791% / 91.843%	0.154004 / 0.182833
unsloth Q4_K_M	0.033473 / 0.033248	90.366% / 91.542%	0.166522 / 0.207902
Our Q4_K_M	0.017288 / 0.017126	92.523% / 93.353%	0.083643 / 0.108646
unsloth UD-Q4_K_XL	0.023854 / 0.024001	91.614% / 92.660%	0.115226 / 0.145981
unsloth Q5_K_S	0.015192 / 0.013366	93.340% / 94.222%	0.072122 / 0.080464
Our Q5_K_M	0.012005 / 0.011123	93.667% / 94.719%	0.058802 / 0.063854

Lower mean KL is not a claim that every token, task or tail metric improves.

What is not better

• Download size is not the same as active-weight size. The pinned bartowski Q4_K_M contains a 15-tensor MTP block with 21,747,712 payload bytes, unused by ordinary decoding. Subtracting that payload from its 579,615,840-byte download gives 557,868,128 bytes, below our Q5_K_M's 569,082,048 bytes. Our files and the tested Unsloth files contain no MTP block. The comparisons above match download sizes, not active-weight budgets.
• Maths: in the historical GSM8K runs every quantized file, ours and public, scored below the original. On the RTX 5090 the original got 145 of 250, our Q4_K_M 137 and Q5_K_M 132, bartowski's Q4_K_M 138 and Unsloth's 131. On the M3 Max the original got 134, our Q5_K_M Mac build 122, bartowski's 129 and Unsloth's 124. Those historical runs supplied no paired GSM8K test. In the October RTX 5090 rerun, Unsloth's Q8_0 (0.812 GB) got 147, the only file above the original, and the public Q5_K_M files got 136 and 137; the paired tests are in the RTX 5090 section. The separate CPU replay above includes paired tests and scoring sensitivities; it does not establish a general maths gain.
• MMLU-Pro: on the M3 Max every file, including the original, got 32 to 39 of 280, and on the RTX 5090 25 to 38 (bartowski's IQ4_XS 8, because it gave no letter on 254 of 280 replies), against about 28 expected from guessing among 10 choices. At this size the test says little about the files.
• Same-label Q5 files: Unsloth's Q5_K_M (0.590 GB) and bartowski's Q5_K_M (0.646 GB, 13.5% larger) are close to our Q5_K_M on KL (.0120 and .0101 against .0098 on WikiText-2) and within a few items of it on GSM8K and MMLU-Pro. On IFEval ours followed 308 prompts against their 295 and 288; no difference on those three tasks survives correction. On tool calls both got more right over the 400 tasks (290 and 282 against 279); on the first 200, Unsloth's got 25 right that ours missed, and ours 4 that it missed. Unsloth's Q8_0 is far closer to the original (KL .0012) and 43% larger than our Q5_K_M.
• Unsloth's UD-Q4_K_XL (0.559 GB) holds up much better on tool calling than the plain Q4_K_M files (104 of 200 selection-exposed on the M3 Max). Our unpublished file at that size recorded KL .0121 against .0215 and tool-calling counts of 112 against 104 (p = 0.23). The tool-calling difference was not significant, so we do not publish it. That decision used the selection-exposed run above.
• The historical tool benchmark covered single-function BFCL simple calls with thinking off. The fresh authored tool and runtime probes have narrower scope than a general task benchmark.
• The RTX 5090 and M3 Max tables use one llama.cpp release (b11617).

Public files: unsloth/Qwen3.5-0.8B-GGUF revision 6ab4614 (Qwen3.5-0.8B-Q4_K_M.gguf; the October rerun added Qwen3.5-0.8B-UD-Q4_K_XL.gguf, Qwen3.5-0.8B-Q5_K_M.gguf and Qwen3.5-0.8B-Q8_0.gguf) and bartowski/Qwen_Qwen3.5-0.8B-GGUF revision f36b1ea (Qwen_Qwen3.5-0.8B-Q4_K_M.gguf; the rerun added Qwen_Qwen3.5-0.8B-IQ4_XS.gguf and Qwen_Qwen3.5-0.8B-Q5_K_M.gguf), each checked by SHA-256. The GPU runs used all layers on the GPU and F16 KV cache.

Image questions

A quick check that the projector works with our files: the first 20 ChartQA test questions (human-written) from the frozen item list of our Qwen3.8 27B release, with lmms-eval's prompt (answer with a single word or phrase), greedy, thinking off, up to 32 output tokens, scored by relaxed accuracy (a number within 5% of the answer, otherwise an exact match ignoring case). llama.cpp b11889 (d854d98) with CUDA on one NVIDIA RTX 5090, -ngl 99 -c 8192 -np 1 --cache-ram 0; Unsloth's file ran with Unsloth's projector.

File	Size	Projector	ChartQA right (of 20)
Artemis Q5_K_M	0.569 GB	ours, F16	11
Artemis Q5_K_M	0.569 GB	ours, BF16	11
Unsloth Q5_K_M	0.590 GB	Unsloth, F16	11

Each file got one question right that the other missed (exact McNemar p = 1.0); 20 questions are too few to separate the files. Our two projectors have the same tensors, byte for byte, as Unsloth's mmproj-F16.gguf and mmproj-BF16.gguf (unsloth/Qwen3.5-0.8B-GGUF revision 6ab4614); only general.* header fields (names, licence, tags) differ. The projectors were converted from Qwen/Qwen3.5-0.8B revision 2fc0636 with convert_hf_to_gguf.py --mmproj from llama.cpp b11889.

File-level fidelity summary

These values repeat the historical RTX 5090 measurements, not the separate CPU replay, on a text used in choosing the files; agreement measures closeness to BF16, not task accuracy.

File	Same next token as BF16, WikiText-2 test
Qwen3.5-0.8B-Artemis-Q5_K_M.gguf	94.2%
Qwen3.5-0.8B-Artemis-Q4_K_M.gguf	92.7%

Conversion and calibration sources

Both files were quantized with llama.cpp release b11617 (commit 14310190c, llama-quantize) from a BF16 conversion of the original made without the MTP layer, using an importance matrix. The calibration text is human-written only:

• Q4_K_M: the first 256 chunks of 512 tokens of the WikiText-2 raw training text (Salesforce/wikitext revision b08601e, wikitext-2-raw-v1, train split). This importance matrix is in the repository as Qwen3.5-0.8B-Artemis-imatrix.gguf.
• Q5_K_M: the whole WikiText-2 raw training text (4,894 chunks of 512 tokens) from the wikitext-2-raw archive. This importance matrix is not published.

WikiText-2 is licensed CC BY-SA 3.0. No WikiText text is included in the files. Both files' headers name TurinTech in general.quantized_by and the calibration set in quantize.imatrix.dataset.

Closeness to the original

The closeness numbers from the tables above, in one fixed layout with the protocol behind each table.

Protocols and standard tables

• Mean KLD: the mean KL divergence between the reference's and the file's next-token distributions, computed and averaged as each protocol's Tool line says. Lower is closer.
• 99th-percentile KLD: the 99th percentile of the per-token values. Lower is closer.
• Same top p: llama.cpp's name for top-1 agreement, the share of scored tokens where the file and the reference pick the same most likely token. Higher is closer.
• PPL and PPL ratio: the file's perplexity on the text, and that perplexity divided by the reference's (1 means the same). They measure prediction of the text, not closeness.

n/a means the card's tables or our run records do not have that value.

Protocol rtx5090-2026-10-20

One NVIDIA RTX 5090, October 2026.

• Reference: BF16 GGUF conversion of Qwen/Qwen3.5-0.8B revision 2fc06364715b967f1860aea9cf38778875588b17, without the MTP layer, made with llama.cpp b11617 (1,516,744,672 bytes, SHA-256 e4cb6d90f531d4ed2b70785683126fb259b7276394d8442be48eaf37cd85e3dd)
• Text wiki: WikiText-2 test; dataset: Salesforce/wikitext at b08601e04326c79dfdd32d625aee71d232d685c3; config: wikitext-2-raw-v1; split: test; chunks: 60; context (tokens): 512; text SHA-256: bbf94c53a05abe9ee670d3b6343608095822c85e26de37c70b24fc571964574a
• Text chat: OpenAssistant validation conversations (English threads as User/Assistant text); dataset: OpenAssistant/oasst1 at fdf72ae0827c1cda404aff25b6603abec9e3399b; split: validation; chunks: 60; context (tokens): 512; text SHA-256: 0a3499a7eaead8f4a40655a868e7faee1ee664e258d09a77cd8f3d97216fc8b7
• Tool: llama-perplexity --kl-divergence against the reference's saved log-probabilities, chunks of 512 tokens; Mean KLD, 99th-percentile KLD (llama.cpp's 99.0% KLD), Same top p and PPL ratio (Mean PPL(Q)/PPL(base)) are llama.cpp's own summary lines
• llama.cpp build: b11617 (commit 14310190cd0a0c831fe949ac8cbedd8abdc90695)
• Command: llama-perplexity -m <file.gguf> -f <text> --kl-divergence-base <reference-text.kld> --kl-divergence -c 512 --chunks 60 -t 16 -ngl 99; the reference logs were written with the same command without --kl-divergence. Taken from the committed scoring script with the chunks, threads and GPU layers recorded for each run
• Hardware: one NVIDIA RTX 5090 (AMD EPYC 7V13 host)
• Settings: 16 threads, all layers on the GPU; the KL command sets no flash attention or KV cache flags, so llama-perplexity's defaults applied (the flash attention and F16 KV settings on the card are for the task runs)
• KL unit: nats (natural log)
• Measured by: TurinTech downloaded the Unsloth and bartowski files at the revisions below, checked each by SHA-256 and measured every file in this table with the same reference, binary and machine
• Selection: the two texts were used in choosing both files, so these columns are not independent tests
• Source on this card: the table columns "KL wiki / chat", "Same next token, wiki / chat"
• Comparison files: Unsloth: unsloth/Qwen3.5-0.8B-GGUF at 6ab461498e20, Qwen3.5-0.8B-Q8_0.gguf, SHA-256 0ad885ffd4bb022fc4f0d33a3308fa108ef8613159d3b3a67e23abca056b7a6c; bartowski: bartowski/Qwen_Qwen3.5-0.8B-GGUF at f36b1ea49a33, Qwen_Qwen3.5-0.8B-Q5_K_M.gguf, SHA-256 7f5b4185c9a3fd54daae8925927377c80e16e3afa094ea9444b3c0a9077136ab; Unsloth: unsloth/Qwen3.5-0.8B-GGUF at 6ab461498e20, Qwen3.5-0.8B-Q5_K_M.gguf, SHA-256 c3ef5827b322c4be08a3a26ce424460d8b37daf593356c2cc1f4e40a7ab0581b; bartowski: bartowski/Qwen_Qwen3.5-0.8B-GGUF at f36b1ea49a33, Qwen_Qwen3.5-0.8B-Q4_K_M.gguf, SHA-256 fb044e93939a70469c905781334f5de1e6c8b608ced6cbc8c9249bd4127d9526; Unsloth: unsloth/Qwen3.5-0.8B-GGUF at 6ab461498e20, Qwen3.5-0.8B-UD-Q4_K_XL.gguf, SHA-256 3177ebd67afe4438374da19e690bc1b98756f7e0fea9240e1be404336156a7b5; Unsloth: unsloth/Qwen3.5-0.8B-GGUF at 6ab461498e20, Qwen3.5-0.8B-Q4_K_M.gguf, SHA-256 bd258782e35f7f458f8aced1adc053e6e92e89bc735ba3be89d38a06121dc517; bartowski: bartowski/Qwen_Qwen3.5-0.8B-GGUF at f36b1ea49a33, Qwen_Qwen3.5-0.8B-IQ4_XS.gguf, SHA-256 0208d8c9491848f83fdd9b125fdcdb7495f4656314f323e5c78da8ae6c5d9426

Protocol rtx5090-2026-10-20, text wiki (WikiText-2 test):

File	Repository	Measured by	Size (GB)	Size (bytes)	Mean KLD (nats)	Same top p (top-1 agreement, %)
BF16 GGUF of Qwen/Qwen3.5-0.8B without MTP	Qwen/Qwen3.5-0.8B at 2fc06364715b	reference	1.52	1516744672	reference	reference
Qwen3.5-0.8B-Artemis-Q5_K_M.gguf	turintech/Qwen3.5-0.8B-Artemis-GGUF at 7408bd7	TurinTech	0.569	569082048	0.0098	94.2
Qwen3.5-0.8B-Artemis-Q4_K_M.gguf	turintech/Qwen3.5-0.8B-Artemis-GGUF at 7408bd7	TurinTech	0.532	532402496	0.0165	92.7
Qwen3.5-0.8B-Q8_0.gguf	unsloth/Qwen3.5-0.8B-GGUF at 6ab461498e20	TurinTech	0.812	811843840	0.0012	98.1
Qwen_Qwen3.5-0.8B-Q5_K_M.gguf	bartowski/Qwen_Qwen3.5-0.8B-GGUF at f36b1ea49a33	TurinTech	0.646	646093920	0.0101	94.4
Qwen3.5-0.8B-Q5_K_M.gguf	unsloth/Qwen3.5-0.8B-GGUF at 6ab461498e20	TurinTech	0.590	590057728	0.0120	94.0
Qwen_Qwen3.5-0.8B-Q4_K_M.gguf	bartowski/Qwen_Qwen3.5-0.8B-GGUF at f36b1ea49a33	TurinTech	0.580	579615840	0.0297	90.8
Qwen3.5-0.8B-UD-Q4_K_XL.gguf	unsloth/Qwen3.5-0.8B-GGUF at 6ab461498e20	TurinTech	0.559	558772480	0.0237	91.8
Qwen3.5-0.8B-Q4_K_M.gguf	unsloth/Qwen3.5-0.8B-GGUF at 6ab461498e20	TurinTech	0.533	532517120	0.0338	90.1
Qwen_Qwen3.5-0.8B-IQ4_XS.gguf	bartowski/Qwen_Qwen3.5-0.8B-GGUF at f36b1ea49a33	TurinTech	0.523	523058272	0.0487	88.7

Protocol rtx5090-2026-10-20, text chat (OpenAssistant validation conversations (English threads as User/Assistant text)):

File	Repository	Measured by	Size (GB)	Size (bytes)	Mean KLD (nats)	Same top p (top-1 agreement, %)
BF16 GGUF of Qwen/Qwen3.5-0.8B without MTP	Qwen/Qwen3.5-0.8B at 2fc06364715b	reference	1.52	1516744672	reference	reference
Qwen3.5-0.8B-Artemis-Q5_K_M.gguf	turintech/Qwen3.5-0.8B-Artemis-GGUF at 7408bd7	TurinTech	0.569	569082048	0.0092	95.1
Qwen3.5-0.8B-Artemis-Q4_K_M.gguf	turintech/Qwen3.5-0.8B-Artemis-GGUF at 7408bd7	TurinTech	0.532	532402496	0.0163	93.6
Qwen3.5-0.8B-Q8_0.gguf	unsloth/Qwen3.5-0.8B-GGUF at 6ab461498e20	TurinTech	0.812	811843840	0.0012	98.2
Qwen_Qwen3.5-0.8B-Q5_K_M.gguf	bartowski/Qwen_Qwen3.5-0.8B-GGUF at f36b1ea49a33	TurinTech	0.646	646093920	0.0096	95.0
Qwen3.5-0.8B-Q5_K_M.gguf	unsloth/Qwen3.5-0.8B-GGUF at 6ab461498e20	TurinTech	0.590	590057728	0.0108	94.8
Qwen_Qwen3.5-0.8B-Q4_K_M.gguf	bartowski/Qwen_Qwen3.5-0.8B-GGUF at f36b1ea49a33	TurinTech	0.580	579615840	0.0295	91.9
Qwen3.5-0.8B-UD-Q4_K_XL.gguf	unsloth/Qwen3.5-0.8B-GGUF at 6ab461498e20	TurinTech	0.559	558772480	0.0238	92.8
Qwen3.5-0.8B-Q4_K_M.gguf	unsloth/Qwen3.5-0.8B-GGUF at 6ab461498e20	TurinTech	0.533	532517120	0.0333	91.4
Qwen_Qwen3.5-0.8B-IQ4_XS.gguf	bartowski/Qwen_Qwen3.5-0.8B-GGUF at f36b1ea49a33	TurinTech	0.523	523058272	0.0474	89.3

Protocol cpu-probe-2026-10-15

Fresh probe, 15 October 2026 (workstation CPU, CPU only, 8 threads).

• Reference: Unsloth's public BF16 GGUF of Qwen/Qwen3.5-0.8B, unsloth/Qwen3.5-0.8B-GGUF at 6ab461498e2023f6e3c1baea90a8f0fe38ab64d0, file Qwen3.5-0.8B-BF16.gguf (1,516,744,736 bytes, SHA-256 cedf89af31c9041b601fa58303285bc46d99c51baee1b13f5e919626ca526ee5)
• Text probe6: Six fixed documents: two recent Wikipedia extracts, two authored code files and two authored chats; chunks: 2 per document; context (tokens): 256; text SHA-256: prose-literature 328e88173fa5ef54cdd72c74ded2968182b4d1e79e89f7dc5ef991c2311740f2; prose-space 872a3d009605ccf22deed05eb9d8eb378d6aad129cd1d26c609f983bdf2a3a53; code-panel 8a86fcd63e889dfffa87a6b5edfe7525f54ea2c2063bb653992e6299e1becd8e; code-runner 41c5af961875d9b5c3e91b82aa3e89c5808408c39293259ee4b402eda07585e1; chat-debug e490eafde70ac0aa94c71bb4b28bc3efaa344d33b4e68becafaf9f022e2eb724; chat-travel 80fbaecbad00bee59eaa569a21d12deea7d6db4635b3e1e887a2e0633ae5102e
• Tool: llama-perplexity --kl-divergence on each of six documents, two chunks of 256 tokens each, against the reference's saved log-probabilities; the value is the plain mean of the six documents' Mean KLD values
• llama.cpp build: b11617 (commit 14310190cd0a0c831fe949ac8cbedd8abdc90695)
• Command: llama-perplexity -m <file.gguf> --kl-divergence-base <reference-document.kld> --kl-divergence -c 256 --chunks 2 -ngl 0 -t 8 for each document; the reference logs were written with llama-perplexity -m <BF16.gguf> -f <document> --save-all-logits <reference-document.kld> -c 256 --chunks 2 -ngl 0 -t 8
• Hardware: workstation CPU, CPU only, 8 threads
• Settings: llama-perplexity defaults otherwise; no flash attention flag
• KL unit: nats (natural log)
• Measured by: TurinTech downloaded the Unsloth and bartowski files at the revisions below, checked each by SHA-256 and measured every file in this table with the same reference, binary and machine
• Selection: the six documents were frozen before the comparisons and were not used for calibration or for choosing files
• Source on this card: the table column "Mean KL ours / public"
• Comparison files: Unsloth: unsloth/Qwen3.5-0.8B-GGUF at 6ab461498e20, Qwen3.5-0.8B-Q4_K_M.gguf, SHA-256 bd258782e35f7f458f8aced1adc053e6e92e89bc735ba3be89d38a06121dc517; bartowski: bartowski/Qwen_Qwen3.5-0.8B-GGUF at f36b1ea49a33, Qwen_Qwen3.5-0.8B-Q4_K_M.gguf, SHA-256 fb044e93939a70469c905781334f5de1e6c8b608ced6cbc8c9249bd4127d9526; Unsloth: unsloth/Qwen3.5-0.8B-GGUF at 6ab461498e20, Qwen3.5-0.8B-UD-Q4_K_XL.gguf, SHA-256 3177ebd67afe4438374da19e690bc1b98756f7e0fea9240e1be404336156a7b5; Unsloth: unsloth/Qwen3.5-0.8B-GGUF at 6ab461498e20, Qwen3.5-0.8B-Q5_K_S.gguf, SHA-256 a201a5899049a7f7845f910ad803574787b4474c3daffc0d33e8688f9114880f

Protocol cpu-probe-2026-10-15, text probe6 (Six fixed documents: two recent Wikipedia extracts, two authored code files and two authored chats):

File	Repository	Measured by	Size (GB)	Size (bytes)	Mean KLD (nats)
Qwen3.5-0.8B-BF16.gguf	unsloth/Qwen3.5-0.8B-GGUF at 6ab461498e20	reference	n/a	1516744736	reference
Qwen3.5-0.8B-Artemis-Q4_K_M.gguf	turintech/Qwen3.5-0.8B-Artemis-GGUF at 12ca0ba7a94a	TurinTech	0.532	532402496	0.022265
Qwen3.5-0.8B-Artemis-Q5_K_M.gguf	turintech/Qwen3.5-0.8B-Artemis-GGUF at 12ca0ba7a94a	TurinTech	0.569	569082048	0.013993
Qwen3.5-0.8B-Q4_K_M.gguf	unsloth/Qwen3.5-0.8B-GGUF at 6ab461498e20	TurinTech	0.533	532517120	0.039005
Qwen_Qwen3.5-0.8B-Q4_K_M.gguf	bartowski/Qwen_Qwen3.5-0.8B-GGUF at f36b1ea49a33	TurinTech	0.580	579615840	0.034045
Qwen3.5-0.8B-UD-Q4_K_XL.gguf	unsloth/Qwen3.5-0.8B-GGUF at 6ab461498e20	TurinTech	0.559	558772480	0.025893
Qwen3.5-0.8B-Q5_K_S.gguf	unsloth/Qwen3.5-0.8B-GGUF at 6ab461498e20	TurinTech	0.569	568889600	0.018341

Protocol cpu-replay-2026-10-16

CPU replay of the original texts, 16 October 2026.

• Reference: Unsloth's public BF16 GGUF of Qwen/Qwen3.5-0.8B, unsloth/Qwen3.5-0.8B-GGUF at 6ab461498e2023f6e3c1baea90a8f0fe38ab64d0, file Qwen3.5-0.8B-BF16.gguf (1,516,744,736 bytes, SHA-256 cedf89af31c9041b601fa58303285bc46d99c51baee1b13f5e919626ca526ee5); the historical producer BF16 used for the RTX 5090 table is a different conversion
• Text wiki: WikiText-2 test; dataset: Salesforce/wikitext at b08601e04326c79dfdd32d625aee71d232d685c3; config: wikitext-2-raw-v1; split: test; chunks: 60; context (tokens): 512; text SHA-256: bbf94c53a05abe9ee670d3b6343608095822c85e26de37c70b24fc571964574a
• Text chat: OpenAssistant validation conversations (English threads as User/Assistant text); dataset: OpenAssistant/oasst1 at fdf72ae0827c1cda404aff25b6603abec9e3399b; split: validation; chunks: 60; context (tokens): 512; text SHA-256: 0a3499a7eaead8f4a40655a868e7faee1ee664e258d09a77cd8f3d97216fc8b7
• Tool: llama-perplexity --kl-divergence against the reference's saved log-probabilities, chunks of 512 tokens; Mean KLD, 99th-percentile KLD (llama.cpp's 99.0% KLD), Same top p and PPL ratio (Mean PPL(Q)/PPL(base)) are llama.cpp's own summary lines
• llama.cpp build: b11617 (commit 14310190c); llama-perplexity SHA-256 1c06240ed8594fd377d655aef2dab0865431e3e779c06638474c96b38e6d74a0
• Command: llama-perplexity -m <file.gguf> --kl-divergence-base <reference-text.kld> --kl-divergence -c 512 --chunks 60 -ngl 0 -t 8
• Hardware: workstation CPU, CPU only, 8 threads
• Settings: llama-perplexity defaults otherwise; no flash attention flag
• KL unit: nats (natural log)
• Measured by: TurinTech downloaded the bartowski and Unsloth files at the revisions below, checked each by SHA-256 and measured every file in this table with the same reference, binary and machine
• Selection: the same two texts were used in choosing the files; they were rebuilt independently to matching SHA-256 hashes for this replay
• Source on this card: the table columns "Wiki / chat P99 token KL", "Wiki / chat mean KL", "Wiki / chat top-1 agreement"
• Comparison files: bartowski: bartowski/Qwen_Qwen3.5-0.8B-GGUF at f36b1ea49a33, Qwen_Qwen3.5-0.8B-Q4_K_M.gguf, SHA-256 fb044e93939a70469c905781334f5de1e6c8b608ced6cbc8c9249bd4127d9526; Unsloth: unsloth/Qwen3.5-0.8B-GGUF at 6ab461498e20, Qwen3.5-0.8B-Q4_K_M.gguf, SHA-256 bd258782e35f7f458f8aced1adc053e6e92e89bc735ba3be89d38a06121dc517; Unsloth: unsloth/Qwen3.5-0.8B-GGUF at 6ab461498e20, Qwen3.5-0.8B-UD-Q4_K_XL.gguf, SHA-256 3177ebd67afe4438374da19e690bc1b98756f7e0fea9240e1be404336156a7b5; Unsloth: unsloth/Qwen3.5-0.8B-GGUF at 6ab461498e20, Qwen3.5-0.8B-Q5_K_S.gguf, SHA-256 a201a5899049a7f7845f910ad803574787b4474c3daffc0d33e8688f9114880f

Protocol cpu-replay-2026-10-16, text wiki (WikiText-2 test):

File	Repository	Measured by	Size (GB)	Size (bytes)	Mean KLD (nats)	99th-percentile KLD (nats)	Same top p (top-1 agreement, %)
Qwen3.5-0.8B-BF16.gguf	unsloth/Qwen3.5-0.8B-GGUF at 6ab461498e20	reference	n/a	1516744736	reference	reference	reference
Qwen_Qwen3.5-0.8B-Q4_K_M.gguf	bartowski/Qwen_Qwen3.5-0.8B-GGUF at f36b1ea49a33	TurinTech	0.580	579615840	0.029953	0.154004	90.791
Qwen3.5-0.8B-Q4_K_M.gguf	unsloth/Qwen3.5-0.8B-GGUF at 6ab461498e20	TurinTech	0.533	532517120	0.033473	0.166522	90.366
Qwen3.5-0.8B-Artemis-Q4_K_M.gguf	turintech/Qwen3.5-0.8B-Artemis-GGUF	TurinTech	0.532	532402496	0.017288	0.083643	92.523
Qwen3.5-0.8B-UD-Q4_K_XL.gguf	unsloth/Qwen3.5-0.8B-GGUF at 6ab461498e20	TurinTech	0.559	558772480	0.023854	0.115226	91.614
Qwen3.5-0.8B-Q5_K_S.gguf	unsloth/Qwen3.5-0.8B-GGUF at 6ab461498e20	TurinTech	0.569	568889600	0.015192	0.072122	93.340
Qwen3.5-0.8B-Artemis-Q5_K_M.gguf	turintech/Qwen3.5-0.8B-Artemis-GGUF	TurinTech	0.569	569082048	0.012005	0.058802	93.667

Protocol cpu-replay-2026-10-16, text chat (OpenAssistant validation conversations (English threads as User/Assistant text)):

File	Repository	Measured by	Size (GB)	Size (bytes)	Mean KLD (nats)	99th-percentile KLD (nats)	Same top p (top-1 agreement, %)
Qwen3.5-0.8B-BF16.gguf	unsloth/Qwen3.5-0.8B-GGUF at 6ab461498e20	reference	n/a	1516744736	reference	reference	reference
Qwen_Qwen3.5-0.8B-Q4_K_M.gguf	bartowski/Qwen_Qwen3.5-0.8B-GGUF at f36b1ea49a33	TurinTech	0.580	579615840	0.029502	0.182833	91.843
Qwen3.5-0.8B-Q4_K_M.gguf	unsloth/Qwen3.5-0.8B-GGUF at 6ab461498e20	TurinTech	0.533	532517120	0.033248	0.207902	91.542
Qwen3.5-0.8B-Artemis-Q4_K_M.gguf	turintech/Qwen3.5-0.8B-Artemis-GGUF	TurinTech	0.532	532402496	0.017126	0.108646	93.353
Qwen3.5-0.8B-UD-Q4_K_XL.gguf	unsloth/Qwen3.5-0.8B-GGUF at 6ab461498e20	TurinTech	0.559	558772480	0.024001	0.145981	92.660
Qwen3.5-0.8B-Q5_K_S.gguf	unsloth/Qwen3.5-0.8B-GGUF at 6ab461498e20	TurinTech	0.569	568889600	0.013366	0.080464	94.222
Qwen3.5-0.8B-Artemis-Q5_K_M.gguf	turintech/Qwen3.5-0.8B-Artemis-GGUF	TurinTech	0.569	569082048	0.011123	0.063854	94.719

File names, sizes and SHA-256 checksums

File names

Bits per weight uses tensor payload bytes, excluding GGUF metadata.

File names end in a standard llama.cpp quant label, so llama.cpp, Ollama and LM Studio can pick a file by tag. Each file is a per-tensor mix of K-quants on a Q4_K base (the Q4_K_M file also has a few small tensors at Q8_0 and F16), with no type making up 90% of the bytes, so the labels are size classes: the smallest standard llama.cpp quant whose file, made from the same BF16 conversion, has at least as many bits per weight. A stock Q4_K_M of this model is 5.51 bits per weight, a stock Q5_K_S 5.88 and a stock Q5_K_M 6.03. The Q5_K_M file (5.93) follows that rule. The Q4_K_M file is an exception: by the rule it would be Q5_K_S, but at 5.54 bits per weight it is 0.6% above a stock Q4_K_M and below Unsloth's Q4_K_M (5.55), the size it was built to stay under, so it keeps the Q4_K_M label. The general.file_type field inside each file matches its label. Public files with the same label can have different tensor allocations. Bartowski's Q4_K_M download also includes an unused MTP block; see the active-weight comparison above.

File	Bits per weight	Main tensor types (share of tensor payload bytes)
Qwen3.5-0.8B-Artemis-Q5_K_M.gguf	5.93	Q5_K 56.1%, Q6_K 43.5%
Qwen3.5-0.8B-Artemis-Q4_K_M.gguf	5.54	Q5_K 72%, Q6_K 15%, Q4_K 11%, Q8_0 1%

File checksums

File	Bytes	sha256
Qwen3.5-0.8B-Artemis-Q5_K_M.gguf	569,082,048	3b782856b66f61b4494b346ad8ff3e91a0d4c294ecedd1797f1855a078abbd05
Qwen3.5-0.8B-Artemis-Q4_K_M.gguf	532,402,496	01e0651f0fc5d0f9a3d1a68166fbdf12ab4d3bb135b3d725d694f42becfaa5a4
Qwen3.5-0.8B-Artemis-imatrix.gguf (importance matrix, not a model)	2,275,008	e0ddf989466e5001798cc5ece4b8ed9f99852ed57e612c5993cb173b29123398
mmproj-BF16.gguf	207,346,272	8fa5933772e64b56d96509b902b70119ce01fd21462f42d8fbb2c8acf0c354a5
mmproj-F16.gguf	204,986,976	07f95999c776a30fc0e2106dc1ad62f110eb745b3f4d7d4de644ec72c19399e3

Artemis

Teams use Artemis to test and improve code against their own objectives. See Artemis.

About and licence

GGUF versions of Qwen3.5 0.8B for llama.cpp, with the vision encoder as a separate projector file. The multi-token prediction (MTP) layer is not included.

Apache-2.0, the licence Qwen releases Qwen3.5 0.8B under (licence); see LICENSE and NOTICE. The importance matrices used for quantization were computed from the WikiText-2 training text (Wikipedia, CC BY-SA 3.0); no WikiText text is included in the files, including the importance-matrix file.

These files keep improving.