PollardWeights/Qwen3-ASR-1.7B-Pollard

Qwen3-ASR-1.7B -- Pollard

Pollard shrank this model: 4.70 GB (f16) -> 1.01 GB -- 79% smaller, 4.7x down.

The smallest rung here; larger, higher-fidelity rungs are listed below.

format	this model's size
f16	4.70 GB
Q8_0	~2.49 GB
Q6_K	1.67 GB
Q4_K_M	~1.36 GB
PollardMix (this repo's IQ2_S)	1.01 GB

Pollard builds of Qwen/Qwen3-ASR-1.7B made with Pollard Weights -- a ladder of measured-allocation quants (bits placed by per-layer sensitivity, not a uniform crush).

Standard GGUF -- every file here runs in stock llama.cpp / ik_llama.cpp, Ollama, LM Studio.

Model details

Parameter count	~2.3B
Architecture	qwen3_asr
Input support	text, audio
imatrix	yes -- see calibration
Perplexity measured	yes -- table below

Which file should I choose?

Every rung is the same weights, sized to a different RAM budget by the measured allocation. Pick the largest one that fits your machine with room for context:

• ~3.7 GB RAM / VRAM -> Q6_K (1.67 GB).
• ~3.3 GB RAM / VRAM -> IQ4_XS (1.28 GB).
• ~3.1 GB RAM / VRAM -> IQ3_S (1.11 GB).
• ~3.0 GB RAM / VRAM -> IQ2_S (1.01 GB).

Available files

file	PPL	size	Mean KLD	Top-1 agree	verified	notes
Qwen3-ASR-1.7B-Pollard-IQ2_S.gguf	7.4685	1.01 GB	0.2996	75.54%	yes	smallest
Qwen3-ASR-1.7B-Pollard-IQ3_S.gguf	6.8277	1.11 GB	0.2121	79.12%	yes
Qwen3-ASR-1.7B-Pollard-IQ4_XS.gguf	5.9772	1.28 GB	0.057	88.34%	yes
Qwen3-ASR-1.7B-Pollard-Q6_K.gguf	5.7427	1.67 GB	0.0174	92.97%	yes	recommended default

Top-1 agree = share of tokens where the rung's most-likely token is the same one the f16 would have picked (llama-perplexity's Same top p for GGUF, the same logit comparison in pollard-verify for the other lanes). Higher is closer to the original model.

Verified = the rung passed pollard-verify --gate: it is the complete model (every tower, projector and processor file of the source), it loads, and it passes the coherence and runtime checks. Its KLD / top-1, tool-call and retention numbers are measured and shown here as they are; "below band" names a number outside the gate file's band for its bit width -- reported, not blocking.

Every format in this repo

One repo holds every Pollard build of this model: the GGUF files above sit at the root, each other format in its own folder: MLX/

MLX (Apple Silicon) -- MLX/

Each rung is a complete checkpoint in its own folder under MLX/. Download only the one you want:

folder	size	avg bits	PPL	Mean KLD	Top-1 agree	Tool-ready	verified
MLX/6bit/	2.24 GB	6.71	--	--	--	--	yes
MLX/5bit/	2.13 GB	6.07	--	--	--	--	yes
MLX/4bit/	2.01 GB	5.43	--	--	--	--	yes
MLX/mlx-4bit/	1.76 GB	4.0	--	--	--	--	yes (below band: mean KLD, top-1 agreement %)
MLX/3bit/	1.70 GB	4.07	--	--	--	--	yes

hf download PollardWeights/Qwen3-ASR-1.7B-Pollard --include "MLX/3bit/*" --local-dir ./Qwen3-ASR-1.7B-Pollard
mlx_lm.server --model ./Qwen3-ASR-1.7B-Pollard/MLX/3bit

Download a specific file

pip install -U "huggingface_hub[cli]"
hf download PollardWeights/Qwen3-ASR-1.7B-Pollard \
  --include "Qwen3-ASR-1.7B-Pollard-IQ2_S.gguf" --local-dir ./

How to run

These are standard GGUF and run with llama.cpp:

llama-server -hf PollardWeights/Qwen3-ASR-1.7B-Pollard:IQ2_S

or from a local file:

llama-cli    -m Qwen3-ASR-1.7B-Pollard-IQ2_S.gguf -ngl 99 -p "Explain why the sky is blue."
llama-server -m Qwen3-ASR-1.7B-Pollard-IQ2_S.gguf -ngl 99      # OpenAI-compatible API + web UI at :8080

They also work in anything built on llama.cpp -- LM Studio, koboldcpp, Jan, ramalama, Ollama (ollama run hf.co/PollardWeights/Qwen3-ASR-1.7B-Pollard).

Use it in your harness

Any agent harness that speaks the OpenAI, Anthropic or Responses API can drive this model through llama-server. Pollard Weights' pollard serve picks the runtime this rung needs, sizes the context to the harness and prints its exact config (--write merges it in, after backing the original up):

pollard serve PollardWeights/Qwen3-ASR-1.7B-Pollard:IQ2_S --harness claude-code      # or omp, codex, opencode, aider, ...

harness	calls	min. context	setup
Oh My Pi (omp)	/v1/chat/completions	32k	~/.omp/agent/models.yml: provider api: openai-completions, baseUrl: http://127.0.0.1:8080/v1, auth: none
Claude Code	/v1/messages	64k	ANTHROPIC_BASE_URL=http://127.0.0.1:8080 ANTHROPIC_AUTH_TOKEN=pollard ANTHROPIC_MODEL=Qwen3-ASR-1.7B-Pollard-IQ2_S
Codex CLI	/v1/responses	32k	~/.codex/config.toml: [model_providers.pollard] base_url = "http://127.0.0.1:8080/v1", wire_api = "responses"
OpenCode	/v1/chat/completions	32k	opencode.json: provider npm: @ai-sdk/openai-compatible, options.baseURL: http://127.0.0.1:8080/v1, limit.context
Cline / Roo Code	/v1/chat/completions	32k	API Provider "OpenAI Compatible", Base URL http://127.0.0.1:8080/v1
Aider	/v1/chat/completions	16k	OPENAI_API_BASE=http://127.0.0.1:8080/v1 aider --model openai/Qwen3-ASR-1.7B-Pollard-IQ2_S
Continue	/v1/chat/completions	32k	config.yaml: provider: llama.cpp, apiBase: http://127.0.0.1:8080
Goose	/v1/chat/completions	32k	GOOSE_PROVIDER=openai OPENAI_HOST=http://127.0.0.1:8080
OpenHands	/v1/chat/completions	32k	model openai/Qwen3-ASR-1.7B-Pollard-IQ2_S, base URL http://host.docker.internal:8080/v1
Open WebUI	/v1/chat/completions	8k	Admin Settings -> Connections -> OpenAI API: http://127.0.0.1:8080/v1
Ollama	own runtime	8k	ollama run hf.co/PollardWeights/Qwen3-ASR-1.7B-Pollard:IQ2_S, then /set parameter num_ctx 8192 or more
LM Studio / Jan	own runtime	8k	load the GGUF, set the context length to 8192 or more

Without Pollard, start llama-server yourself with a context the harness fits in, the model's own template (tool calls are parsed through it) and the thinking kept out of the answer:

llama-server -hf PollardWeights/Qwen3-ASR-1.7B-Pollard:IQ2_S -c 65536 --jinja --reasoning-format deepseek --alias Qwen3-ASR-1.7B-Pollard-IQ2_S -np 1 --port 8080

Too small a context fails silently: the harness's system prompt and tool schemas are truncated and the model looks like it ignores its tools. Claude Code needs 64k.

imatrix (calibration)

The importance matrix (Qwen3-ASR-1.7B.imatrix, included) was computed on the pollard-calib 'general' profile (sources, revisions and licences in the shipped .sources.json).

The exact corpus is included as release-calib.txt, so the allocation can be reproduced rather than taken on trust.

ARM / AVX

llama.cpp repacks weights into an interleaved layout at load time for faster inference on ARM and AVX machines -- no special file needed, online repacking covers these quants. The old Q4_0_4_4/4_8/8_8 variants are not required.

Errata

• Every file here loads in stock llama.cpp -- verified from the tensor types with pollard-ggufcheck, not assumed from the filenames.
• Measured allocation places bits by per-layer sensitivity under a size budget.
• Single machine; replication invited.

Credits & license

• Base model: Qwen/Qwen3-ASR-1.7B (Qwen)
• Quantization tooling: llama.cpp (ggml-org)
• Method + tooling: Pollard Weights -- measure first, no claim before a number.
• License: apache-2.0, inherited from the base model.

Built with Pollard Weights -- frontier models, small hardware, no compromise.