Update README and more changes, rebase to main (#1259)

huggingface · Aug 16, 2024 · 41e0a3f · 41e0a3f
1 parent 09551d9
commit 41e0a3f
Show file tree

Hide file tree

Showing 3 changed files with 12 additions and 27 deletions.
diff --git a/examples/contrastive-image-text/README.md b/examples/contrastive-image-text/README.md
@@ -118,7 +118,8 @@ python run_clip.py \
     --gaudi_config_name Habana/clip \
     --throughput_warmup_steps 3 \
     --dataloader_num_workers 16 \
-    --bf16
+    --bf16 \
+    --trust_remote_code
 ```
 
 
@@ -151,7 +152,8 @@ python ../gaudi_spawn.py --world_size 8 --use_mpi run_clip.py \
     --mediapipe_dataloader \
     --use_hpu_graphs_for_training \
     --bf16 \
-    --distribution_strategy fast_ddp
+    --distribution_strategy fast_ddp \
+    --trust_remote_code
 ```
 
 > `--mediapipe_dataloader` only works on Gaudi2.
@@ -182,7 +184,8 @@ python ../gaudi_spawn.py --world_size 8 --use_deepspeed run_clip.py \
     --use_hpu_graphs_for_inference \
     --gaudi_config_name Habana/clip \
     --throughput_warmup_steps 3 \
-    --deepspeed path_to_my_deepspeed_config
+    --deepspeed path_to_my_deepspeed_config \
+    --trust_remote_code
 ```
 
 You can look at the [documentation](https://huggingface.co/docs/optimum/habana/usage_guides/deepspeed) for more information about how to use DeepSpeed in Optimum Habana.
@@ -231,7 +234,8 @@ python ../gaudi_spawn.py --use_mpi --world_size 8 run_bridgetower.py \
   --logging_steps 10 \
   --dataloader_num_workers 1 \
   --mediapipe_dataloader \
-  --distribution_strategy fast_ddp
+  --distribution_strategy fast_ddp \
+  --trust_remote_code
 ```
 
 > `--mediapipe_dataloader` only works on Gaudi2.
@@ -260,7 +264,8 @@ python run_clip.py \
     --use_hpu_graphs_for_inference \
     --gaudi_config_name Habana/clip \
     --bf16 \
-    --mediapipe_dataloader
+    --mediapipe_dataloader \
+    --trust_remote_code
 ```
 
 > `--mediapipe_dataloader` only works on Gaudi2.
diff --git a/examples/text-generation/README.md b/examples/text-generation/README.md
@@ -215,24 +215,6 @@ python run_generation.py \
 > The prompt length is limited to 16 tokens. Prompts longer than this will be truncated.
 
 
-### Use PEFT models for generation
-
-You can also provide the path to a PEFT model to perform generation with the argument `--peft_model`.
-
-For example:
-```bash
-python run_generation.py \
---model_name_or_path meta-llama/Llama-2-7b-hf \
---use_hpu_graphs \
---use_kv_cache \
---batch_size 1 \
---bf16 \
---max_new_tokens 100 \
---prompt "Here is my prompt" \
---peft_model goliaro/llama-2-7b-lora-full
-```
-
-
 ### Using growing bucket optimization
 
 With `--bucket_size`, instead of padding up the kv-cache up to full size before starting, we grow the cache/input in multiples of `bucket_size`. This helps increase throughput and also reduce number of compilations if the dataset has varying prompt lengths.
@@ -603,8 +585,6 @@ deepspeed --num_gpus 8 run_lm_eval.py \
 -o eval.json
 ```
 
-> If the dataset you want to use requires the execution of remote code, please set the following environment variable: `HF_DATASETS_TRUST_REMOTE_CODE=true`
-
 
 ## Text-Generation Pipeline
 

diff --git a/examples/visual-question-answering/README.md b/examples/visual-question-answering/README.md
@@ -63,6 +63,6 @@ python run_openclip_vqa.py \
 ```
 
 Models that have been validated:
-  - [BiomedCLIP-PubMedBERT_256-vit_base_patch16_224](https://huggingface.co/microsoft/BiomedCLIP-PubMedBERT_256-vit_base_patch16_224)
+  - [microsoft/BiomedCLIP-PubMedBERT_256-vit_base_patch16_224](https://huggingface.co/microsoft/BiomedCLIP-PubMedBERT_256-vit_base_patch16_224)
   - [laion/CLIP-ViT-g-14-laion2B-s12B-b42K](https://huggingface.co/laion/CLIP-ViT-g-14-laion2B-s12B-b42K)
-  - [apple/DFN5B-CLIP-ViT-H-14](https://huggingface.co/apple/DFN5B-CLIP-ViT-H-14/tree/main)
+  - [apple/DFN5B-CLIP-ViT-H-14](https://huggingface.co/apple/DFN5B-CLIP-ViT-H-14/tree/main)