Recipe
step-3-7-flash-q4-k-s-ryzen-ai-max-plus-395-128gb-llama-cpp-tp1
step-3-7-flash-q4-k-s-ryzen-ai-max-plus-395-128gb-llama-cpp-tp1Observed LocalMaxxing leaderboard run. Evidence for compatibility, not an executable launch contract.
Record
- Status
- candidate
- Source
- localmaxxing
- Engine
- llama.cpp
- Engine version
- toolbox-rocm-7.2.4
- Accelerators
- 1
- Tensor parallel
- 1
- Context tokens
- 65,536
- chat
- unknown
- reasoning
- unknown
- tools
- unknown
- vision
- unknown
Hugging Face model card
Identity
https://huggingface.co/stepfun-ai/Step-3.7-Flash-GGUF- Repository
- stepfun-ai/Step-3.7-Flash-GGUF
- 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
llama.cpp
Evidence only · candidate · reference
Candidate evidence — not a Run contract
No tokenized launch fields. This is measured or documented compatibility, not a Docker launch.
Measured speed
| Concurrency | Context | Prefill | Decode | TTFT ms | Status | Sweep |
|---|---|---|---|---|---|---|
| 1 | 65,536 | 234.6 | 22 | 36,084.1 | observed | step-3-7-flash-q4-k-s-ryzen-ai-max-plus-395-128gb-llama-cpp-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
- step-3-7-flash-q4-k-s-ryzen-ai-max-plus-395-128gb-llama-cpp-tp1
- model instance id
- stepfun-ai-step-3-7-flash-gguf--q4-k-s
- recipe source
- localmaxxing
- schema version
- local-ai-registry/v1
- speed sweep ids
- step-3-7-flash-q4-k-s-ryzen-ai-max-plus-395-128gb-llama-cpp-tp1-sweep
- status
- candidate
capabilities
engine
- name
- llama.cpp
- version
- toolbox-rocm-7.2.4
serving
- max context tokens
- 65,536
- 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
- 2026-08-31T23:03:15Z
serving.max concurrency · reason server-capacity-not-evidenced
serving.max concurrency · state unknown
metadata
- localmaxxing · backend
- rocm
- localmaxxing · batch size
- 1
- localmaxxing · hardware label
- Ryzen AI Max 395
- localmaxxing · notes
- Horizon Step-3.7-Flash Q4_K_S benchmark. 198B MoE VL model (~11B active). BIOS UMA=1GB; kernel cmdline amd_iommu=off amdgpu.gttsize=126976 ttm.pages_limit=32505856. ROCm sees Radeon 8060S as 126976 MiB; GTT 124 GiB. Measured via kyuz0/amd-strix-halo-toolboxes:rocm-7.2.4 on port 8020. Scenario: single streamed chat-completion, ctx=65536, target ~8k content tokens, max_tokens=64, temperature=0. tokSPrefill = native llama.cpp metrics delta (prompt_tokens_total / prompt_seconds_total). Run-specific nonce + cold prompt prevents KV cache reuse. Prompt delta verified (8460). Includes mmproj-step3.7-flash-f16.gguf for vision support. Server flags: -ngl 999 -fa 1 --no-mmap --jinja --metrics --reasoning-format auto --no-webui. No MTP/speculative flags. Local artifact: /home/joshua/benchmarks/step-3.7-flash-q4ks-2026-06-29/run-8k.json
- localmaxxing · run id
- cmqznm4yn02rmoe01nv85v1t0
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/cmqznm4yn02rmoe01nv85v1t0 ↗ |