Recipe
qwen3-5-0-8b-base-mq4-rx-7900-xtx-24gb-hipfire-tp1
qwen3-5-0-8b-base-mq4-rx-7900-xtx-24gb-hipfire-tp1Observed LocalMaxxing leaderboard run. Evidence for compatibility, not an executable launch contract.
Record
- Status
- candidate
- Source
- localmaxxing
- Engine
- hipfire
- Engine version
- 0.3.0+44eb828cc804
- Accelerators
- 1
- Tensor parallel
- 1
- Context tokens
- 128
- Max concurrency
- 1
- chat
- unknown
- reasoning
- unknown
- tools
- unknown
- vision
- unknown
Hugging Face model card
Identity
https://huggingface.co/Qwen/Qwen3.5-0.8B- Repository
- Qwen/Qwen3.5-0.8B
- 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
Observed source tokens
Mechanical split of the source command. Unverified against the engine CLI. The registry does not offer Run for this recipe.
scripts/lmx_redline_campaign.py--model/home/kaden/.hipfire/models/qwen3.5-0.8b.registry.mq4--prompt-filebenchmarks/prompts/sweep/cap.txt--device0--process-runs3--bench-runs5--transportpm4--kv-modeq8--context128--iterations100--max-seq2048
| Flag | Value |
|---|---|
--model | /home/kaden/.hipfire/models/qwen3.5-0.8b.registry.mq4 |
--prompt-file | benchmarks/prompts/sweep/cap.txt |
--device | 0 |
--process-runs | 3 |
--bench-runs | 5 |
--transport | pm4 |
--kv-mode | q8 |
--context | 128 |
--iterations | 100 |
--max-seq | 2048 |
Measured speed
| Concurrency | Context | Prefill | Decode | TTFT ms | Status | Sweep |
|---|---|---|---|---|---|---|
| 1 | 128 | — | 550.2 | 69 | observed | qwen3-5-0-8b-base-mq4-rx-7900-xtx-24gb-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
- rx-7900-xtx-24gb
- id
- qwen3-5-0-8b-base-mq4-rx-7900-xtx-24gb-hipfire-tp1
- model instance id
- qwen-qwen3-5-0-8b--mq4
- recipe source
- localmaxxing
- schema version
- local-ai-registry/v1
- speed sweep ids
- qwen3-5-0-8b-base-mq4-rx-7900-xtx-24gb-hipfire-tp1-sweep
- status
- candidate
capabilities
engine
- name
- hipfire
- version
- 0.3.0+44eb828cc804
serving
- max concurrency
- 1
- max context tokens
- 128
- 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
metadata
- localmaxxing · backend
- rocm
- localmaxxing · hardware label
- RX 7900 XTX
- localmaxxing · notes
- SINGLE-STREAM. hipfire Redline retained-PM4 vs plain HIP A/B on identical MQ4 weights, 3 fresh processes x5 measurement rows, campaign medians. PM4 550.162 tok/s vs HIP 505.823 tok/s, speedup 1.088x (PM4 wins). engine: hipfire 0.3.0+44eb828cc804, backend rocm, gfx1100 AMD Radeon RX 7900 XTX quant: MQ4 = MagnumQuant 4-bit group-256 FWHT-rotated, 0.53125 B/weight =4.25 bpw ideal, file 549751424 B =>5.04 bpw effective on 0.87B total params (5.50 bpw on 0.80B text) model file: qwen3.5-0.8b.registry.mq4 sha256 aedfe31be68213dbcd13cff3a63ba150c330d0d0547df30f2c629e76ab5dda39 prompt: cap.txt md5 bbf2d0483ecacaa2eb7296f9c8083f95, 13 prompt tokens (est.), context 128, iterations 100 method: median of 3 runs, fresh process, temperature 0 (greedy), PM4 retained replay verified at positions 128/227 each process evidence: sealed case qwen08-mq4-redline-pm4-20260809-gfx1100-rx7900xtx, corpus manifest 44eb828cc804 not measured: peak VRAM (not instrumented this campaign) TTFT 69.0 ms is the median of the same campaign's coherence pass (3 rows, auto arm, same binary/model/route); tok/s is the bench median. Separate passes, both measured.
- localmaxxing · observed command
- scripts/lmx_redline_campaign.py --model /home/kaden/.hipfire/models/qwen3.5-0.8b.registry.mq4 --prompt-file benchmarks/prompts/sweep/cap.txt --device 0 --process-runs 3 --bench-runs 5 --transport pm4 --kv-mode q8 --context 128 --iterations 100 --max-seq 2048
- localmaxxing · run id
- cmsnp33j900n9o001ieskp9t0
- localmaxxing · tokenized · arguments
- scripts/lmx_redline_campaign.py, --model, /home/kaden/.hipfire/models/qwen3.5-0.8b.registry.mq4, --prompt-file, benchmarks/prompts/sweep/cap.txt, --device, 0, --process-runs, 3, --bench-runs, 5, --transport, pm4, --kv-mode, q8, --context, 128, --iterations, 100, --max-seq, 2048
- localmaxxing · tokenized · fidelity
- faithful
provenance
- captured at
- 2026-08-30T09:10:02Z
sources
| captured at | kind | url |
|---|---|---|
| 2026-08-30T09:10:02Z | normalized-recipe | www.localmaxxing.com/en/runs/cmsnp33j900n9o001ieskp9t0 ↗ |