FreshContextA context lab · Fife
Drawing
02 · Benchmarks
Sheet
2 of 3
Rev
A · 2026-10-01

MiniCPM5-2B · file← MiniCPM5-2B

BF16 GGUF

A full-size copy, made only as the starting point for the two ternary copies. Never served; gone. Converted here.

Size
5.0 GB
Status
in a scratch folder; gone (not on record when)
Scored runs
0
ResultsNever scored

This file never completed a scored run. Its record below says why.

The recordHow it was made · how it was served · what went wrong · what happened to it

Rebuilt from the lab's records by Claude, an AI model. The results above come straight from the automated runs. How these pages are made.

Filea scratch file
FormatBF16 GGUF, 5.03 GB, 381 tensors
Where frommade here 2026-09-17 with convert_hf_to_gguf.py (llama.cpp-stq tree)
Statusin a scratch folder; gone (not on record when)

How it was made

  • Input: models/minicpm5-2b-hf (the 2026-09-17 BF16 safetensors download).
  • Command, run from llama.cpp-stq:
uv run --with numpy --with sentencepiece --with transformers --with torch --with gguf \
python convert_hf_to_gguf.py models/minicpm5-2b-hf --outfile minicpm5-2b-bf16.gguf --outtype bf16
  • Tool: convert_hf_to_gguf.py from the llama.cpp-stq tree (upstream 0cea36222 plus the AngelSlim STQ1_0 patch). The gguf-py version is not on record.
  • Date: 2026-09-17 09:59Z. Output: 381 tensors, 5.03 GB; "4800.66 MiB (16.00 BPW)" per llama-quantize.
  • imatrix made from it: llama-imatrix -m minicpm5-2b-bf16.gguf -f calib.txt -o minicpm5-2b.imatrix --chunks 64 -ngl 999. The calibration text was the llama.cpp-stq README and docs (209 KB), chosen so no private text was baked into weights. Final PPL 6.3822 ± 0.13870.
  • Who asked: part of the ternary test.

How it was served

labeldates (UTC)binary and versionctx / -np / KV / -fa / --jinja / reasoningharness profilewhy this label exists
— (none)————never served; conversion and imatrix input only

Never served; no run-log label.

Decisions

Made as the full-precision input for STQ1_0 and PTQ1_0 (reconstructed: the log never says why OpenBMB's F16 GGUF was not used).

What went wrong

None recorded.

What happened to it

Made 2026-09-17 in a scratch folder → used for imatrix, STQ1_0, PTQ1_0 → gone; when is not on record (the scratch folder is not swept by any RESTORE script). No RESTORE entry. Recreate with the command above from openbmb/MiniCPM5-2B.