Tweaks

Signed-off-by: Prabhu Subramanian <[email protected]>
CycloneDX · Feb 7, 2025 · 4d5d4b7 · 4d5d4b7
1 parent 685ac70
commit 4d5d4b7
Show file tree

Hide file tree

Showing 2 changed files with 12 additions and 11 deletions.
diff --git a/contrib/fine-tuning/fine-tune-mlx.sh b/contrib/fine-tuning/fine-tune-mlx.sh
@@ -21,26 +21,27 @@ node prepare.js dataset
 node validator.js dataset
 
 echo "Test base model with the prompt 'Tell me about cdxgen'. Usually yields a low-quality response."
-mlx_lm.generate --model ./${BASE_MODEL} --prompt "Tell me about cdxgen" --temp 0.05
+mlx_lm.generate --model ${BASE_MODEL} --prompt "Tell me about cdxgen" --temp 0.05
 
 # We first convert from HF to mlx
 rm -rf ${BASE_MODEL_MLX}
 mlx_lm.convert --hf-path ${BASE_MODEL} --mlx-path ${BASE_MODEL_MLX}
 
 echo "Weight-Decomposed Low-Rank Adaptation (DoRA) fine-tuning ${BASE_MODEL_MLX} with cdx1 dataset. This might take a while ..."
-mlx_lm.lora --model ${BASE_MODEL_MLX} --train --data dataset --fine-tune-type dora --batch-size 1 --num-layers ${NUM_LAYERS} --iters 1000 --grad-checkpoint
+mlx_lm.lora --model ${BASE_MODEL_MLX} --train --data dataset --fine-tune-type lora --batch-size 1 --num-layers ${NUM_LAYERS} --iters 1000 --grad-checkpoint
 
 echo "Fuse model to ${FUSED_MODEL} using the cdx1 adapters"
 rm -rf ${FUSED_MODEL} ${FUSED_GGUF_MODEL}
-mlx_lm.fuse --model ${BASE_MODEL_MLX} --adapter-path adapters --hf-path ${FUSED_MODEL} --save-path ${FUSED_MODEL} --de-quantize --export-gguf --gguf-path cdx1-bf16.gguf
+mlx_lm.fuse --model ${BASE_MODEL_MLX} --adapter-path adapters --hf-path ${FUSED_MODEL} --save-path ${FUSED_MODEL} --de-quantize --export-gguf --gguf-path cdx1-f16.gguf
 
 echo "Test fused model with the prompt 'Tell me about cdxgen'. Must yield a better response."
 mlx_lm.generate --model ./${FUSED_MODEL} --prompt "Tell me about cdxgen" --temp 0.05
 
 mkdir -p ${FUSED_GGUF_MODEL}
-mv ${FUSED_MODEL}/cdx1-bf16.gguf ${FUSED_GGUF_MODEL}
+mv ${FUSED_MODEL}/cdx1-f16.gguf ${FUSED_GGUF_MODEL}
 cp Modelfile ${FUSED_GGUF_MODEL}/
 cp ${FUSED_MODEL}/*.json ${FUSED_MODEL}/merges.txt ${FUSED_GGUF_MODEL}/
+sed -i'' 's|CycloneDX/cdx1-gguf|./cdx1-f16.gguf|g' ${FUSED_GGUF_MODEL}/Modelfile
 
 echo "Create quantized models"
 rm -rf ${QUANT_MODEL_8BIT}

diff --git a/contrib/fine-tuning/upload-hf.sh b/contrib/fine-tuning/upload-hf.sh
@@ -13,13 +13,13 @@ QUANT_MODEL_4BIT=${HF_ORG}/${TOOL_BASE_MODEL}-${TUNING_TOOL}-4bit
 huggingface-cli whoami
 
 echo "Uploading datasets"
-huggingface-cli upload --repo-type dataset CycloneDX/cdx-docs ./cdxgen-docs cdxgen-docs
-huggingface-cli upload --repo-type dataset CycloneDX/cdx-docs ./guides guides
+huggingface-cli upload --quiet --repo-type dataset CycloneDX/cdx-docs ./cdxgen-docs cdxgen-docs
+huggingface-cli upload --quiet --repo-type dataset CycloneDX/cdx-docs ./guides guides
 
 echo "Uploading models. Please wait ..."
-huggingface-cli upload --repo-type model ${QUANT_MODEL_8BIT} ./${QUANT_MODEL_8BIT} .
-huggingface-cli upload --repo-type model ${QUANT_MODEL_6BIT} ./${QUANT_MODEL_6BIT} .
-huggingface-cli upload --repo-type model ${QUANT_MODEL_4BIT} ./${QUANT_MODEL_4BIT} .
+huggingface-cli upload --quiet --repo-type model ${QUANT_MODEL_8BIT} ./${QUANT_MODEL_8BIT} .
+huggingface-cli upload --quiet --repo-type model ${QUANT_MODEL_6BIT} ./${QUANT_MODEL_6BIT} .
+huggingface-cli upload --quiet --repo-type model ${QUANT_MODEL_4BIT} ./${QUANT_MODEL_4BIT} .
 
-huggingface-cli upload --repo-type model ${FUSED_MODEL} ./${FUSED_MODEL} .
-huggingface-cli upload --repo-type model ${FUSED_GGUF_MODEL} ./${FUSED_GGUF_MODEL} .
+huggingface-cli upload --quiet --repo-type model ${FUSED_MODEL} ./${FUSED_MODEL} .
+huggingface-cli upload --quiet --repo-type model ${FUSED_GGUF_MODEL} ./${FUSED_GGUF_MODEL} .