Update README.md

README.md

@@ -2,8 +2,14 @@
 license: cc-by-sa-4.0
 datasets:
 - bigcode/the-stack-dedup
+- sahil2801/CodeAlpaca-20k
+- teknium/GPTeacher-CodeInstruct
+model-base:
+- replit/replit-code-v1-3b
 tags:
 - code
+- instruct
+- self instruct
 language:
 - code
 programming_language: 
@@ -27,152 +33,52 @@ programming_language:
 - Jupyter Notebook
 - R
 - Shell
-model-index:
-- name: replit-code-v1-3b
-  results:
-  - task: 
-      name: Code Generation
-      type: code-generation
-    dataset:
-      name: "HumanEval" 
-      type: openai_humaneval
-    metrics:
-    - name: pass@1
-      type: pass@1
-      value: 0.219
-      verified: false
 ---
 
+Base Model: replit/replit-code-v1-3b
 
-# replit-code-v1-3b
-Developed by: Replit, Inc.
+This model is fine tuned on both Sahil2801's CodeAlpaca & Teknium's GPTeacher Code-Instruct to give Replit's Code model instruct capabilities.
 
-[**🧑‍💻 Test it on our Demo Space! 🧑‍💻**](https://huggingface.co/spaces/replit/replit-code-v1-3b-demo)
+Dataset links:
+CodeAlpaca: https://huggingface.co/datasets/sahil2801/CodeAlpaca-20k
+GPTeacher subset - Code Instruct: https://github.com/teknium1/GPTeacher
 
-## Model Description
-`replit-code-v1-3b` is a 2.7B Causal Language Model focused on **Code Completion**. The model has been trained on a subset of the [Stack Dedup v1.2 dataset](https://arxiv.org/abs/2211.15533).
+This model was trained on 2x a100 80gb for 1 hour on ~25,000 code instruction/response pairs in Alpaca format.
 
-The training mixture includes **20 different languages**, listed here in descending order of number of tokens: 
-<br/>
-`Markdown`, `Java`, `JavaScript`, `Python`, `TypeScript`, `PHP`, `SQL`, `JSX`, `reStructuredText`, `Rust`, `C`, `CSS`, `Go`, `C++`, `HTML`, `Vue`, `Ruby`, `Jupyter Notebook`, `R`, `Shell`
-<br/>
-In total, the training dataset contains 175B tokens, which were repeated over 3 epochs -- in total, `replit-code-v1-3b` has been trained on **525B** tokens (~195 tokens per parameter).
+Refer to the base models HuggingFace model card for some basic requirements to run: https://huggingface.co/replit/replit-code-v1-3b
 
-The model has been trained on the [MosaicML](https://www.mosaicml.com/) platform with 256 x A100-40GB GPUs, leveraging their latest [LLM examples repo](https://github.com/mosaicml/examples/tree/release/v0.0.4/examples/llm).
-<br/>
-`replit-code-v1-3b` is powered by state-of-the-art LLM techniques, such as: 
-[Flash Attention](https://arxiv.org/abs/2205.14135) for fast training and inference,
-[AliBi positional embeddings](https://arxiv.org/abs/2108.12409) to support variable context length at inference time, 
-[LionW optimizer](https://arxiv.org/abs/2302.06675), 
-etc.
-
-## Intended Use
-Replit intends this model be used by anyone as a foundational model for application-specific fine-tuning without strict limitations on commercial use.
-
-## Limitations
-The pre-training dataset may have contained offensive or inappropriate content even after applying data cleansing filters, and such content may be reflected in model generated text. We recommend that users exercise reasonable caution when using in production systems. Do not use for any applications that may cause harm or distress to individuals or groups.
-
-## License
-The model checkpoint and vocabulary file are licensed under the Creative Commons license (CC BY-SA-4.0). Under the license, you must give credit to Replit, provide a link to the license, and indicate if changes were made. You may do so in any reasonable manner, but not in any way that suggests that Replit endorses you or your use.
-
-The source code files (`*.py`) are licensed under the Apache 2.0 license.
-
-## Contact
-For questions and comments about the model, please post in the community section. 
-
-## How to Use
-First of all, you need to install the latest versions of the following dependencies:
-```
-einops
-sentencepiece
-torch
-transformers
+This fine tune can be prompted like any alpaca fine tune:
 ```
+### Instruction:
+<prompt>
 
-You can then load the model as follows:
-```python
-from transformers import AutoModelForCausalLM
+### Input:
+<additional context>
 
-# load model
-model = AutoModelForCausalLM.from_pretrained('replit/replit-code-v1-3b', trust_remote_code=True)
+### Response:
 ```
 
-To use the optimized Triton implementation of FlashAttention on GPUs with BF16 precision, first install the following dependencies: 
-```
-flash-attn==0.2.8
-triton==2.0.0.dev20221202
-```
-
-Then, move the model to `bfloat16` and use it as follows:
-```python
-from transformers import AutoModelForCausalLM
-
-# load model
-model = AutoModelForCausalLM.from_pretrained('replit/replit-code-v1-3b', trust_remote_code=True, attn_impl='triton')
-model.to(device='cuda:0', dtype=torch.bfloat16)
-
-# forward pass
-x = torch.tensor([[0, 1, 2, 3, 4, 5, 6, 7, 8, 9]])
-x = x.to(device='cuda:0')
-y = model(x)
+or  
 
 ```
+### Instruction:
+<prompt>
 
-Note that `trust_remote_code=True` is passed to the `from_pretrained` method because ReplitLM is not a class in the
-[Transformers](https://huggingface.co/docs/transformers/index) library. 
-
-### Tokenizer
-
-We have trained a custom SentencePiece Unigram tokenizer optimized with a vocabulary specifically for code of 32768 tokens.
+### Response:
 
-Note that using this requires the `sentencepiece` library to be installed. 
-
-The tokenizer can be used as follows:
-
-```python
-from transformers import AutoTokenizer
-
-# load tokenizer
-tokenizer = AutoTokenizer.from_pretrained('replit/replit-code-v1-3b', trust_remote_code=True)
-
-# single input encoding + generation
-x = tokenizer.encode('def hello():\n  print("hello world")\n', return_tensors='pt')
-y = model.generate(x)
-
-# decoding, clean_up_tokenization_spaces=False to ensure syntactical correctness
-generated_code = tokenizer.decode(y[0], skip_special_tokens=True, clean_up_tokenization_spaces=False)
-print(generated_code)
 ```
 
-Note that: 
-- `trust_remote_code=True` is passed to the `from_pretrained` method because ReplitLM is not a class in the [Transformers](https://huggingface.co/docs/transformers/index) library. 
-- `clean_up_tokenization_spaces=False` is meant to avoid removing spaces in the output, because that would affect the syntactical correctness of the generated code. 
-
-
-### Generation
+This model for me produced coherent outputs with the following sampler settings, but feel free to experiment:  
+```  
+max_new_tokens=128, do_sample=True, use_cache=True, temperature=0.2, top_p=0.9, eos_token_id= self.tokenizer.eos_token_id
+```  
 
-You can generate code using the `transformers` library as follows:
+In the tokenizer decode arguments, it also needs these settings:  
+```  
+skip_special_tokens=True, clean_up_tokenization_space=False
+```  
 
-```python
-from transformers import AutoModelForCausalLM, AutoTokenizer
-
-tokenizer = AutoTokenizer.from_pretrained('replit/replit-code-v1-3b', trust_remote_code=True)
-model = AutoModelForCausalLM.from_pretrained('replit/replit-code-v1-3b', trust_remote_code=True)
-
-x = tokenizer.encode('def fibonacci(n): ', return_tensors='pt')
-y = model.generate(x, max_length=100, do_sample=True, top_p=0.95, top_k=4, temperature=0.2, num_return_sequences=1, eos_token_id=tokenizer.eos_token_id)
-
-# decoding, clean_up_tokenization_spaces=False to ensure syntactical correctness
-generated_code = tokenizer.decode(y[0], skip_special_tokens=True, clean_up_tokenization_spaces=False)
-print(generated_code)
+The following parameters were used with HuggingFace trainer to train the model with:  
 ```
-
-Experiment with different decoding methods and parameters to get the best results for your use case.
-
-### Post Processing
-
-Note that as with all code generation models, post-processing of the generated code is important. In particular, the following post-processing steps are recommended:
-- stop generation when the EOS token is encountered
-- remove trailing whitespaces
-- set `max_tokens` to a reasonable value based on your completion use case
-- truncate generation to stop words such as `return`, `def`, "```", "`\n\n\n`" to avoid generating incomplete code when `max_tokens` is larger than the length of the expected generated code.
+--model_name_or_path replit/replit-code-v1-3b --data_path /root/stanford_alpaca/train.json --bf16 True --output_dir /root/stanford_alpaca/model_ckpts --num_train_epochs 3 --per_device_train_batch_size 4 --per_device_eval_batch_size 1 --gradient_accumulation_steps 8 --save_strategy steps --save_steps 200 --save_total_limit 3 --learning_rate 1e-5 --weight_decay 0. --warmup_ratio 0.03 --tf32 True --run_name Replit1
+```