Update README.md

README.md

@@ -12,8 +12,7 @@ datasets:
 ## Model Description
 
 LLaMA-2-7B-32K-Chat is an open-source, long-context chat model finetuned from [LLaMA-2-7B-32K](https://huggingface.co/togethercomputer/LLaMA-2-7B-32K), over high-quality instruction and chat data.
-We built Llama-2-7B-32K-Chat with less than 200 lines of Python script using [Together API](https://together.ai/blog/api-announcement), and we also make the recipe fully available.
-For more details, please refer to our [Github repo](https://github.com/togethercomputer/LLaMA-2-32K-Chat).
+We built Llama-2-7B-32K-Chat with less than 200 lines of Python script using [Together API](https://together.ai/blog/api-announcement), and we also make the [recipe fully available](https://github.com/togethercomputer/LLaMA-2-32K-Chat).
 We hope that this can enable everyone to finetune their own version of [LLaMA-2-7B-32K](https://huggingface.co/togethercomputer/LLaMA-2-7B-32K) — play with [Together API](https://together.ai/blog/api-announcement) and give us feedback! 
 
 ## Data Collection Details
@@ -24,12 +23,10 @@ LLaMA-2-7B-32K-Chat is fine-tuned over a combination of two parts:
    The complete dataset is also released [here](https://huggingface.co/datasets/togethercomputer/llama-instruct).
    We also share the complete recipe for the data collection process [here](https://github.com/togethercomputer/LLaMA-2-32K-Chat).
    
-3. **4K instructions of summarization from the BookSum datasets**.
-   BookSum is a unique dataset designed to address the challenges of long-form narrative summarization.
-   This dataset features source documents from the literature domain, including novels, plays, and stories, and offers human-written, highly abstractive summaries.
-   We here focus on chapter-level data. BookSum poses a unique set of challenges, necessitating that the model comprehensively read through each chapter.
-   We used 4K of the instructions in our fine-tuning.
+2. **Long-context Summarization and Long-context QA**.
+   We follow the recipe of [LLaMA-2-7B-32K](https://together.ai/blog/llama-2-7b-32k), and train our model with the [BookSum dataset](https://huggingface.co/datasets/togethercomputer/Long-Data-Collections) and [Multi-document Question Answering](https://arxiv.org/abs/2307.03172).
 
+The final data mixture used for model finetuning is: 19K instruction (50%) + BookSum (25%) + MQA (25%).
 
 ## Model Usage
 
@@ -41,6 +38,9 @@ from transformers import AutoTokenizer, AutoModelForCausalLM
 
 tokenizer = AutoTokenizer.from_pretrained("togethercomputer/LLaMA-2-7B-32K-Chat")
 model = AutoModelForCausalLM.from_pretrained("togethercomputer/LLaMA-2-7B-32K-Chat", trust_remote_code=True, torch_dtype=torch.float16)
+input_ids = tokenizer.encode(<your instruction>, return_tensors="pt")
+output = model.generate(input_ids, max_length=..., temperature=...)
+output_text = tokenizer.decode(output[0], skip_special_tokens=True)
 ```
 
 The model is also hosted on [Together Playground](https://api.together.xyz/playground). You can simply play with the model by using prompt formatted by:
@@ -82,15 +82,29 @@ Their charm, a gift, that's forever told.
 
 ## Model Evaluation
 
-We evaluate the model with [PG19 dataset](https://huggingface.co/datasets/pg19) and compare the perplexity with [Llama-2-7b-chat](https://huggingface.co/meta-llama/Llama-2-7b-chat-hf),
-the results are summarized below (note that the perplexity is normalized following the protocol [here](https://together.ai/blog/llama-2-7b-32k)).
+We evaluate the model from three aspects: 1) [Normalized perplexity](https://together.ai/blog/llama-2-7b-32k) over [PG19 dataset](https://huggingface.co/datasets/pg19);
+2) [Rouge score over BookSum](https://together.ai/blog/llama-2-7b-32k); and
+3) [Accuracy over Multi-document Question Answering (MQA)](https://together.ai/blog/llama-2-7b-32k). We summarize the results below:
 
+* Normalized Perplexity over PG19
 | Model | 2K Seq | 4K Seq | 8K Seq | 16K Seq | 32K Seq |
 | -------- | ------- | ------- | ------- | ------- | ------- |
 | LLaMA-2-7B-Chat (Meta) | 1.844 | 1.833 | N/A | N/A | N/A |
 | LLaMA-2-7B-32K-Chat (ours) | 1.813 | 1.798 | 1.781 | 1.778 | 1.772|
 
-We observe that LLaMA-2-7B-32K-Chat obtains reasonable (and even better) perplexity, comparable to the original LLaMA-2-7B-Chat model.
+* Rouge Score over BookSum
+| Model | R1 | R2 | RL |
+| -------- | ------- | ------- | ------- |
+| LLaMA-2-7B-Chat (Meta) | 0.055 | 0.008 | 0.046 |
+| LLaMA-2-7B-32K-Chat (ours) | 0.365 | 0.086 | 0.192 |
+
+* Accuracy over MQA
+| Model | 20 docs (Avg 2.9K tokens) | 30 docs (Avg 4.4K tokens) | 50 docs (Avg 7.4K tokens) |
+| -------- | ------- | ------- | ------- |
+| LLaMA-2-7B-Chat (Meta) | 0.384 | 0.375 | 0.313 |
+| LLaMA-2-7B-32K-Chat (ours) | 0.451 | 0.434 | 0.373 |
+
+We observe that LLaMA-2-7B-32K-Chat obtains reasonable (and even better) perplexity, rouge score and accuracy over the original LLaMA-2-7B-Chat model.
 
 ## Limitations and Bias