Recipe

qwen3-6-35b-a3b-mq4r-radeon-ai-pro-r9700-32gb-hipfire-tp1

qwen3-6-35b-a3b-mq4r-radeon-ai-pro-r9700-32gb-hipfire-tp1

Observed LocalMaxxing leaderboard run. Evidence for compatibility, not an executable launch contract.

Record

Status
candidate
Source
localmaxxing
Engine
hipfire
Engine version
0.3.0+b0bcc3f91506
Accelerators
1
Tensor parallel
1
chat
unknown
reasoning
unknown
tools
unknown
vision
unknown

Hugging Face model card

Identity

https://huggingface.co/Qwen/Qwen3.6-35B-A3B
Repository
Qwen/Qwen3.6-35B-A3B
Status
known
Link type
Exact Hub repository

Public Hugging Face repository confirmed by the Hub API.

Candidate: useful compatibility or speed evidence. The registry does not offer Run until promotion requirements are met.

Observed configuration

hipfire

Evidence only · candidate · reference

Candidate evidence — not a Run contract

Source
https://www.localmaxxing.com/en/runs/cmsnsxbrr02fbo0012bo2krxv

Observed source tokens

Mechanical split of the source command. Unverified against the engine CLI. The registry does not offer Run for this recipe.

  1. python3
  2. scripts/lmx_continuous_batch.py
  3. --model
  4. qwen3.6-35b-a3b.mq4r
  5. --prompt-file
  6. benchmarks/prompts/merge_sort_thinking_off.txt
  7. --batch-size
  8. 64
  9. --requests
  10. 64
  11. --runs
  12. 3
  13. --max-tokens
  14. 128
  15. --max-seq
  16. 4096
  17. --device
  18. 1
FlagValue
--modelqwen3.6-35b-a3b.mq4r
--prompt-filebenchmarks/prompts/merge_sort_thinking_off.txt
--batch-size64
--requests64
--runs3
--max-tokens128
--max-seq4096
--device1

Measured speed

ConcurrencyContextPrefillDecodeTTFT msStatusSweep
644,09617.4628.5observedqwen3-6-35b-a3b-mq4r-radeon-ai-pro-r9700-32gb-hipfire-tp1-sweep

Remaining fields

Identity, launch, related records, and measured speed are shown above. This is the rest of the normalized record.

hardware count
1
hardware id
radeon-ai-pro-r9700-32gb
id
qwen3-6-35b-a3b-mq4r-radeon-ai-pro-r9700-32gb-hipfire-tp1
model instance id
qwen-qwen3-6-35b-a3b--mq4r
recipe source
localmaxxing
schema version
local-ai-registry/v1
speed sweep ids
qwen3-6-35b-a3b-mq4r-radeon-ai-pro-r9700-32gb-hipfire-tp1-sweep
status
candidate

capabilities

engine

name
hipfire
version
0.3.0+b0bcc3f91506

serving

tensor parallel
1
Provenance & metadata (3)

facts

capabilities.chat · provenance · captured at
2026-08-30T09:10:02Z
capabilities.chat · reason capability-not-verified
capabilities.chat · state unknown
capabilities.reasoning · provenance · captured at
2026-08-30T09:10:02Z
capabilities.reasoning · reason capability-not-verified
capabilities.reasoning · state unknown
capabilities.tools · provenance · captured at
2026-08-30T09:10:02Z
capabilities.tools · reason capability-not-verified
capabilities.tools · state unknown
capabilities.vision · provenance · captured at
2026-08-30T09:10:02Z
capabilities.vision · reason capability-not-verified
capabilities.vision · state unknown
engine.graph mode · provenance · captured at
2026-08-30T09:10:02Z
engine.graph mode · reason runtime-detail-not-published
engine.graph mode · state unknown
serving.kv cache tokens · provenance · captured at
2026-08-30T09:10:02Z
serving.kv cache tokens · reason kv-cache-capacity-not-published
serving.kv cache tokens · state unknown
serving.max concurrency · provenance · captured at
2026-09-01T10:20:39Z
serving.max concurrency · reason server-capacity-not-evidenced
serving.max concurrency · state unknown
serving.max context tokens · provenance · captured at
2026-09-01T10:20:39Z
serving.max context tokens · reason context-limit-not-evidenced
serving.max context tokens · state unknown

metadata

localmaxxing · backend
rocm
localmaxxing · batch size
64
localmaxxing · hardware label
Radeon AI Pro R9700
localmaxxing · notes
BATCHED AGGREGATE THROUGHPUT — not single-stream decode. batch=64, concurrency=64, 64 requests/run, 3 runs, continuous batching, max_tokens=128, KV=q8, greedy; TTFT measured under load. single R9700 (1 of 4 in the box; other cards ran unrelated jobs and were unused). engine: hipfire 0.3.0+b0bcc3f91506, backend rocm, gfx1201 quant: MQ4R = MagnumQuant 4-bit Redline (MQ4 attention/router/shared + graded MQ4 routed experts; '.mq4r' also selects Redline retained-PM4 dispatch), 4.16 bpw effective. WORKLOAD (read before comparing): prompt merge_sort_thinking_off.txt md5 253c7ac50857fe6d0e10fb0d2c5e35c0, only 38 prompt tokens, 128 generated. The contextLength 4096 on this row is the max_seq ALLOCATION, not the prompt size. The Arc Pro B70 row on this page (1139.8 tok/s, concurrency 64) reports contextLength 16384, but that is likewise its --max-model-len; its own notes state diverse 512-token prompts. Its per-request prefill work is therefore ~13x ours, so this row is NOT like-for-like and should not be read as matching or beating it on equal workload. method: median of 3 runs; samples 1113.68 / 1115.74 / 1131.49 tok/s; per-lane 17.43 tok/s; median latency 7182.37 ms. All outputs byte-identical and coherent. capacity: 64 lanes is the ceiling at max_seq 4096 on 32 GB; B96 and B128 both fail with 'hipMalloc: out of memory / continuous batch allocation failed'. evidence: 20260810T221534Z-a3b-mq4r-concurrency-ladder not measured: peak VRAM.
localmaxxing · observed command
python3 scripts/lmx_continuous_batch.py --model qwen3.6-35b-a3b.mq4r --prompt-file benchmarks/prompts/merge_sort_thinking_off.txt --batch-size 64 --requests 64 --runs 3 --max-tokens 128 --max-seq 4096 --device 1
localmaxxing · run id
cmsnsxbrr02fbo0012bo2krxv
localmaxxing · tokenized · arguments
python3, scripts/lmx_continuous_batch.py, --model, qwen3.6-35b-a3b.mq4r, --prompt-file, benchmarks/prompts/merge_sort_thinking_off.txt, --batch-size, 64, --requests, 64, --runs, 3, --max-tokens, 128, --max-seq, 4096, --device, 1
localmaxxing · tokenized · fidelity
faithful

provenance

captured at
2026-08-30T09:10:02Z

sources

captured atkindurl
2026-08-30T09:10:02Znormalized-recipewww.localmaxxing.com/en/runs/cmsnsxbrr02fbo0012bo2krxv