Hugging Face's logo

mradermacher
/
model_requests

English
Model card Files
xet
 Community
2348

revit-cpder

#2314
by kuklev9797 - opened
Discussion
kuklev9797

Hi mradermacher,

Thank you for quantizing schauh11/revit-coder-14b β€” your GGUFs are
the only public artifact of this model right now.

The original BF16 safetensors shards (model-0000X-of-00006) have been
removed from huggingface.co/schauh11/revit-coder-14b β€” only the config
and tokenizer remain. The Wayback Machine has no archive of them either.

I am trying to use this model for production fine-tuning experiments,
but I am encountering quality issues with the quantized versions
(both Q4_K_M and Q8_0 produce broken/random output on simple prompts
like "What is 2+2?", while base Qwen3-14B works correctly on the same
Ollama installation).

Would you happen to still have:

  1. The original BF16 source weights cached locally from when you
    ran the quantization on 2026-03-03? Even one of the six shards
    would be helpful for diagnosing whether the issue is with the
    quantization process or with the upstream model itself.

  2. Or, if storage is an issue, could you possibly run a quick
    verification: load your Q8_0 with llama.cpp and prompt
    "Hello, what is 2+2?" β€” does it produce coherent text on your
    end, or random tokens?

If the source weights are gone but you have time for a re-quant from
imatrix or from your nico1 worker logs, I would be very grateful β€”
or even just confirmation that the upstream weights were already
broken at conversion time would save me weeks of debugging.

I'm happy to upload the broken-quantization repro to a HF Space if
that helps.

Thank you for the immense service you do for the open-weights
community β€” your 30,000+ quants are an indispensable resource.

Best regards,
DIma
RichardErkhov

  1. Hi, sorry, we are not keeping original bf16 source weights, as that would mean storing petabytes of models, which is... impossible, as we are sadly not millionaires, just a few guys with shared ideas an a few computers =( =)

>llama-cli.exe --model revit-coder-14b.Q8_0.gguf
load_backend: loaded RPC backend from C:\Users\Administrator\Downloads\llama-b8827-bin-win-cuda-13.1-x64\ggml-rpc.dll
load_backend: loaded CPU backend from C:\Users\Administrator\Downloads\llama-b8827-bin-win-cuda-13.1-x64\ggml-cpu-alderlake.dll

Loading model...


β–„β–„ β–„β–„
β–ˆβ–ˆ β–ˆβ–ˆ
β–ˆβ–ˆ β–ˆβ–ˆ  β–€β–€β–ˆβ–„ β–ˆβ–ˆβ–ˆβ–„β–ˆβ–ˆβ–ˆβ–„  β–€β–€β–ˆβ–„    β–„β–ˆβ–ˆβ–ˆβ–ˆ β–ˆβ–ˆβ–ˆβ–ˆβ–„ β–ˆβ–ˆβ–ˆβ–ˆβ–„
β–ˆβ–ˆ β–ˆβ–ˆ β–„β–ˆβ–€β–ˆβ–ˆ β–ˆβ–ˆ β–ˆβ–ˆ β–ˆβ–ˆ β–„β–ˆβ–€β–ˆβ–ˆ    β–ˆβ–ˆ    β–ˆβ–ˆ β–ˆβ–ˆ β–ˆβ–ˆ β–ˆβ–ˆ
β–ˆβ–ˆ β–ˆβ–ˆ β–€β–ˆβ–„β–ˆβ–ˆ β–ˆβ–ˆ β–ˆβ–ˆ β–ˆβ–ˆ β–€β–ˆβ–„β–ˆβ–ˆ β–ˆβ–ˆ β–€β–ˆβ–ˆβ–ˆβ–ˆ β–ˆβ–ˆβ–ˆβ–ˆβ–€ β–ˆβ–ˆβ–ˆβ–ˆβ–€
                                    β–ˆβ–ˆ    β–ˆβ–ˆ
                                    β–€β–€    β–€β–€

build      : b8827-5e6c0e18b
model      : revit-coder-14b.Q8_0.gguf
modalities : text

available commands:
  /exit or Ctrl+C     stop or exit
  /regen              regenerate the last response
  /clear              clear the chat history
  /read <file>        add a text file
  /glob <pattern>     add text files using globbing pattern


> Hello, what is 2+2?

Wran1CarStimesζˆ—cancelunIlluminate]
)
)
)),)

mathrmar]
):
),rightarrow).

sd]
]
(xTdnenddBnTn****

Sign up or log in to comment