Recipe
granite-4-2-8b-q4-k-s-rtx-3060-12gb-llama-cpp-tp1
granite-4-2-8b-q4-k-s-rtx-3060-12gb-llama-cpp-tp1Observed LocalMaxxing leaderboard run. Evidence for compatibility, not an executable launch contract.
Record
- Status
- candidate
- Source
- localmaxxing
- Engine
- llama.cpp
- Engine version
- 0.1.2-dev (build 197, commit b062ba735)
- Accelerators
- 1
- Tensor parallel
- 1
- Context tokens
- 4,096
- Max concurrency
- 1
- chat
- unknown
- reasoning
- unknown
- tools
- unknown
- vision
- unknown
Hugging Face model card
Identity
https://huggingface.co/ibm-granite/granite-4.2-8b-GGUF- Repository
- ibm-granite/granite-4.2-8b-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
Observed source tokens
Mechanical split of the source command. Unverified against the engine CLI. The registry does not offer Run for this recipe.
llama-quantize--imatriximatrix-granite42-mixed.gguf--token-embedding-typeq5_kbf16.ggufout.ggufQ4_K_S;bench:llama-bench-p512-n128-ngl99-faon-r3
| Flag | Value |
|---|---|
--imatrix | imatrix-granite42-mixed.gguf |
--token-embedding-type | q5_k |
-p | 512 |
-n | 128 |
-ngl | 99 |
-fa | on |
-r | 3 |
Measured speed
| Concurrency | Context | Prefill | Decode | TTFT ms | Status | Sweep |
|---|---|---|---|---|---|---|
| 1 | 4,096 | 1,748 | 57.1 | 312 | observed | granite-4-2-8b-q4-k-s-rtx-3060-12gb-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
- rtx-3060-12gb
- id
- granite-4-2-8b-q4-k-s-rtx-3060-12gb-llama-cpp-tp1
- model instance id
- ibm-granite-granite-4-2-8b-gguf--q4-k-s
- recipe source
- localmaxxing
- schema version
- local-ai-registry/v1
- speed sweep ids
- granite-4-2-8b-q4-k-s-rtx-3060-12gb-llama-cpp-tp1-sweep
- status
- candidate
capabilities
draft launch
- accelerator backend
- nvidia
- arguments
- -hf, ibm-granite/granite-4.2-8b-GGUF:Q4_K_S, --n-gpu-layers, 999, --host, 0.0.0.0, --port, 8080, -c, 4096
- container port
- 8,080
- environment · LLAMA CACHE
- /root/.cache/huggingface
- host port
- 8,080
- image
- ghcr.io/ggml-org/llama.cpp:server-cuda12-b10481@sha256:b2497f8834f5ecb4e38530f6bf2734b8e0be107f0f0857e259672d1cb85b71c2
- ipc
- host
- kind
- docker
- shm size
- 16g
- synthesized · generated at
- 2026-08-31T22:12:17Z
- synthesized · image provenance
- gemma-4-12b-q4-k-m-rtx-3060-12gb-llama-cpp-tp1
- synthesized · template
- llama-cpp-server-v1
mounts
| read only | target |
|---|---|
| No | /root/.cache/huggingface |
engine
- name
- llama.cpp
- version
- 0.1.2-dev (build 197, commit b062ba735)
serving
- max concurrency
- 1
- max context tokens
- 4,096
- 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
- cuda
- localmaxxing · hardware label
- RTX 3060
- localmaxxing · notes
- Custom quant 'Q4KS-XL': Q4_K_S base + token_embd upgraded to Q5_K, quantized from bf16 (no requant) with custom importance matrix calibrated on ~600K mixed tokens (Wikipedia wikitext-103 prose + local C++/TS/Python code), 40L GQA dense granite arch, untied embeddings. Measured vs official Q4_K_M same GPU/build: KLD vs bf16 0.1333 vs 0.1796 (-26%), same-top-p 88.1% vs 86.0%, tg128 57.06 vs 54.24 (+5.2%), pp512 1748 vs 1724, file 4.79 GiB vs 5.10 GiB. Strictly better than stock Q4_K_M on all axes. r=3 medians.
- localmaxxing · observed command
- llama-quantize --imatrix imatrix-granite42-mixed.gguf --token-embedding-type q5_k bf16.gguf out.gguf Q4_K_S; bench: llama-bench -p 512 -n 128 -ngl 99 -fa on -r 3
- localmaxxing · run id
- cmt8zl9b300ikli01o2v42b0q
- localmaxxing · tokenized · arguments
- llama-quantize, --imatrix, imatrix-granite42-mixed.gguf, --token-embedding-type, q5_k, bf16.gguf, out.gguf, Q4_K_S;, bench:, llama-bench, -p, 512, -n, 128, -ngl, 99, -fa, on, -r, 3
- 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/cmt8zl9b300ikli01o2v42b0q ↗ |