Recipe
deepseek-v4-flash-mq2-lloyd-ryzen-ai-max-plus-395-128gb-hipfire-tp1
deepseek-v4-flash-mq2-lloyd-ryzen-ai-max-plus-395-128gb-hipfire-tp1Observed LocalMaxxing leaderboard run. Evidence for compatibility, not an executable launch contract.
Record
- Status
- candidate
- Source
- localmaxxing
- Engine
- hipfire
- Engine version
- v0.2.1+5d3683a
- Accelerators
- 1
- Tensor parallel
- 1
- Context tokens
- 16,384
- Max concurrency
- 1
- chat
- unknown
- reasoning
- unknown
- tools
- unknown
- vision
- unknown
Hugging Face model card
Identity
https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash- Repository
- deepseek-ai/DeepSeek-V4-Flash
- 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.
hipfireservedeepseek-v4-flash0.0.0.0:11435--kv-modeq8
Environment
| Variable | Value |
|---|---|
HIPFIRE_DEEPSEEK4_SPEC_DECODE | 1 |
HIPFIRE_DEEPSEEK4_SPEC_K | 2 |
Measured speed
| Concurrency | Context | Prefill | Decode | TTFT ms | Status | Sweep |
|---|---|---|---|---|---|---|
| 1 | 16,384 | 144.5 | 15.2 | — | observed | deepseek-v4-flash-mq2-lloyd-ryzen-ai-max-plus-395-128gb-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
- ryzen-ai-max-plus-395-128gb
- id
- deepseek-v4-flash-mq2-lloyd-ryzen-ai-max-plus-395-128gb-hipfire-tp1
- model instance id
- deepseek-ai-deepseek-v4-flash--mq2-lloyd
- recipe source
- localmaxxing
- schema version
- local-ai-registry/v1
- speed sweep ids
- deepseek-v4-flash-mq2-lloyd-ryzen-ai-max-plus-395-128gb-hipfire-tp1-sweep
- status
- candidate
capabilities
engine
- name
- hipfire
- version
- v0.2.1+5d3683a
serving
- max concurrency
- 1
- max context tokens
- 16,384
- 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
- Ryzen AI Max 395
- localmaxxing · notes
- hipfire (Rust RDNA-native engine, Kaden-Schutt/hipfire rev 5d3683a) on Framework Desktop Strix Halo gfx1151, 120G GTT, ROCm 7.2.4. nwoolmer/hipfire-deepseek-v4-flash MQ2-Lloyd mixed precision (86.2G, 2-bit MagnumQuant-Lloyd routed experts, Q8F16/F16 elsewhere) + 2.0G MTP sidecar, Q8 KV cache, spec decode K=2 (K=2 beat K=1/3/4 in sweep). Measured on diverse code-review prompt, streaming timing: pp1887 depth: 15.2 tok/s decode / 144.5 prefill. Depth-honest long-context: pp15317: 13.2 decode / 140.7 prefill. Spec decode is load-bearing: no-spec decode is 4.1 tok/s.
- localmaxxing · observed command
- HIPFIRE_DEEPSEEK4_SPEC_DECODE=1 HIPFIRE_DEEPSEEK4_SPEC_K=2 hipfire serve deepseek-v4-flash 0.0.0.0:11435 --kv-mode q8
- localmaxxing · run id
- cmrh7bdda009smj0170x1khux
- localmaxxing · tokenized · arguments
- hipfire, serve, deepseek-v4-flash, 0.0.0.0:11435, --kv-mode, q8
- localmaxxing · tokenized · environment · HIPFIRE DEEPSEEK4 SPEC DECODE
- 1
- localmaxxing · tokenized · environment · HIPFIRE DEEPSEEK4 SPEC K
- 2
- 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/cmrh7bdda009smj0170x1khux ↗ |