Edit model card

llama.cpp conversion of https://huggingface.co/euclaise/Memphis-scribe-3B

except for f16 and q8_0, every quant is using the imatrix from wiki-train

ppl (512 wiki.test)

quant ppl (lower is better)
f16(baseline) 9.9437 +/- 0.07019
q8_0 9.9474 +/- 0.07022
q5_k_m 10.0347 +/- 0.07091
q4_k_m 10.1192 +/- 0.07152
iq3_xxs 11.5266 +/- 0.08157
q2_k 13.0623 +/- 0.09548
iq2_xs 16.6174 +/- 0.11807
iq2_xxs 22.6462 +/- 0.16226
Downloads last month
189
GGUF
Model size
2.8B params
Architecture
stablelm
Unable to determine this model’s pipeline type. Check the docs .

Quantized from