CoolFace
Modelpublic

RichardErkhov/smcleish_-_clrs_gemma_2b_100k_finetune_with_traces-gguf

sourceHugging Faceupdated 2y agoView on Hugging Face
0likes337downloads
Model Card

Quantization made by Richard Erkhov.

Github

Discord

Request more models

clrsgemma2b100kfinetunewithtraces - GGUF

  • —Model creator: https://huggingface.co/smcleish/
  • —Original model: https://huggingface.co/smcleish/clrsgemma2b100kfinetunewithtraces/
NameQuant methodSize
clrs_gemma_2b_100k_finetune_with_traces.Q2_K.ggufQ2_K1.08GB
clrs_gemma_2b_100k_finetune_with_traces.IQ3_XS.ggufIQ3_XS1.16GB
clrs_gemma_2b_100k_finetune_with_traces.IQ3_S.ggufIQ3_S1.2GB
clrs_gemma_2b_100k_finetune_with_traces.Q3_K_S.ggufQ3KS1.2GB
clrs_gemma_2b_100k_finetune_with_traces.IQ3_M.ggufIQ3_M1.22GB
clrs_gemma_2b_100k_finetune_with_traces.Q3_K.ggufQ3_K1.29GB
clrs_gemma_2b_100k_finetune_with_traces.Q3_K_M.ggufQ3KM1.29GB
clrs_gemma_2b_100k_finetune_with_traces.Q3_K_L.ggufQ3KL1.36GB
clrs_gemma_2b_100k_finetune_with_traces.IQ4_XS.ggufIQ4_XS1.4GB
clrs_gemma_2b_100k_finetune_with_traces.Q4_0.ggufQ4_01.44GB
clrs_gemma_2b_100k_finetune_with_traces.IQ4_NL.ggufIQ4_NL1.45GB
clrs_gemma_2b_100k_finetune_with_traces.Q4_K_S.ggufQ4KS1.45GB
clrs_gemma_2b_100k_finetune_with_traces.Q4_K.ggufQ4_K1.52GB
clrs_gemma_2b_100k_finetune_with_traces.Q4_K_M.ggufQ4KM1.52GB
clrs_gemma_2b_100k_finetune_with_traces.Q4_1.ggufQ4_11.56GB
clrs_gemma_2b_100k_finetune_with_traces.Q5_0.ggufQ5_01.68GB
clrs_gemma_2b_100k_finetune_with_traces.Q5_K_S.ggufQ5KS1.68GB
clrs_gemma_2b_100k_finetune_with_traces.Q5_K.ggufQ5_K1.71GB
clrs_gemma_2b_100k_finetune_with_traces.Q5_K_M.ggufQ5KM1.71GB
clrs_gemma_2b_100k_finetune_with_traces.Q5_1.ggufQ5_11.79GB
clrs_gemma_2b_100k_finetune_with_traces.Q6_K.ggufQ6_K1.92GB
clrs_gemma_2b_100k_finetune_with_traces.Q8_0.ggufQ8_02.49GB

Original model description: --- libraryname: transformers license: mit basemodel:

  • —google/gemma-2b ---

Model Details

google/gemma-2b model finetuned on 100,000 CLRS-Text examples.

Training Details

  • —Learning Rate: 1e-4, 150 warmup steps then cosine decayed to 5e-06 using AdamW optimiser
  • —Batch size: 128
  • —Loss taken over answer only, not on question.