README.md

---
license: apache-2.0
language:
- en
pipeline_tag: text-generation
library_name: transformers
widget:
- example_title: Write python function and explain
  text: >-
    <s> <|sys_start|> You are an AI assistant. You will be given a task. You
    must generate a detailed and long answer. <|sys_end|> <|im_start|> Write a
    python function that takes a list of integers and returns the squared sum of
    the list. <|im_end|>
  output:
    text: "\n\nHere's a Python function named `squared_sum_list` that takes a list of integers as input and returns the squared sum of the list:\n\n```python\ndef squared_sum_list(lst):\n\treturn sum([num ** 2 for num in lst])\n```\nThe function `squared_sum_list` uses a list comprehension to iterate over each number in the input list `lst` and calculate its square. Then, it uses the `sum` function to accumulate all the squared numbers in a single value - the squared sum of the list.\n\nFor example:\n\n```python\nnumbers = [1, 2, 3, 4, 5]\nprint(squared_sum_list(numbers))  # Outputs: 55\n```\n\nIn the above code, the list `[1, 2, 3, 4, 5]` is passed as an argument to the `squared_sum_list` function. The function calculates the sum of the squares of the elements in the list, which is `1 + 4 + 9 + 16 + 25 = 55`. The function then returns this result, which is printed to the console."
tags:
- llm
- code
datasets:
- openaccess-ai-collective/oasst1-guanaco-extended-sharegpt
- Open-Orca/SlimOrca
- AtAndDev/ShareGPT-Vicuna-v3-cleaned-unfiltered
- WizardLM/WizardLM_evol_instruct_V2_196k
- winglian/chatlogs-en-cleaned
- HuggingFaceH4/CodeAlpaca_20K
- theblackcat102/evol-codealpaca-v1
- nickrosh/Evol-Instruct-Code-80k-v1
- open-phi/textbooks
- open-phi/programming_books_llama
- LLM360/CrystalCoderDatasets
model-index:
- name: CrystalChat
  results:
  - task:
      type: text-generation
    dataset:
      type: openai_humanneval
      name: OpenAI HumanEval
    metrics:
    - name: pass@1 (t=0.01)
      type: pass@1
      value: 31.707
    - name: pass@10 (t=0.8)
      type: pass@10
      value: 65.755
  - task:
      type: text-generation
    dataset:
      type: mbpp
      name: Mostly Basic Python Problems (mbpp)
    metrics:
    - name: pass@1 (t=0.01)
      type: pass@1
      value: 39.4
    - name: pass@10 (t=0.8)
      type: pass@10
      value: 59.895
  - task:
      type: multiple-choice
    dataset:
      type: race
      name: RACE
    metrics:
    - name: accuracy
      type: accuracy
      value: 41.148
  - task:
      type: multiple-choice
    dataset:
      type: mmlu
      name: Measuring Massive Multitask Language Understanding (MMLU)
    metrics:
    - name: accuracy
      type: accuracy
      value: 52.789
  - task:
      type: multiple-choice
    dataset:
      type: truthful_qa
      name: Truthful QA
    metrics:
    - name: accuracy
      type: accuracy
      value: 47.29
  - task:
      type: multiple-choice
    dataset:
      type: winogrande
      name: Winogrande
    metrics:
    - name: accuracy (5 shot)
      type: accuracy
      value: 70.639
    - name: accuracy (0 shot)
      type: accuracy
      value: 68.114
  - task:
      type: multiple-choice
    dataset:
      type: copa
      name: COPA
    metrics:
    - name: accuracy
      type: accuracy
      value: 85
  - task:
      type: text-classification
    dataset:
      type: boolq
      name: Boolq
    metrics:
    - name: accuracy
      type: accuracy
      value: 82.783
  - task:
      type: question-answering
    dataset:
      type: openbookqa
      name: Openbook QA
    metrics:
    - name: accuracy
      type: accuracy
      value: 42
  - task:
      type: multiple-choice
    dataset:
      type: hellaSwag
      name: HellaSwag
    metrics:
    - name: accuracy (10-shot)
      type: accuracy
      value: 76.12
    - name: accuracy (0-shot)
      type: accuracy
      value: 73.312
  - task:
      type: question-answering
    dataset:
      type: piqa
      name: PIQA
    metrics:
    - name: accuracy
      type: accuracy
      value: 77.856
  - task:
      type: question-answering
    dataset:
      type: ai2_arc
      name: ARC (Easy)
    metrics:
    - name: accuracy
      type: accuracy
      value: 70.328
  - task:
      type: question-answering
    dataset:
      type: ai2_arc
      name: ARC (Challenge)
    metrics:
    - name: accuracy (25-shot)
      type: accuracy
      value: 51.706
    - name: accuracy (0-shot)
      type: accuracy
      value: 44.625
  - task:
      type: text-generation
    dataset:
      type: gsm8k
      name: GSM8K (Grade School Math 8K)
    metrics:
    - name: Accuracy (5 shot)
      type: accuracy
      value: 28.052
---

# CrystalChat

We present CrystalChat, an instruction following model finetuned from [LLM360/CrystalCoder](https://huggingface.co/LLM360/CrystalCoder). Following the release of [LLM360/AmberChat](https://huggingface.co/LLM360/AmberChat) and [LLM360/AmberSafe](https://huggingface.co/LLM360/AmberSafe) in December 2023, CrystalChat is the next and most performant chat model released under LLM360. CrystalChat is trained on a carefully selected mix publicly available language and code datasets.

As always, the training data, training code, and metrics are publicly available.

## About LLM360

LLM360 is an initiative for comprehensive and fully open-sourced LLMs, 
where all training details, model checkpoints, intermediate results, and 
additional analyses are made available to the community. Our goal is to advance 
the field by inviting the community to deepen the understanding of LLMs 
together. As the first step of the project LLM360, we release all intermediate 
model checkpoints, our fully-prepared pre-training dataset, all source code and
configurations, and training details. We are
committed to continually pushing the boundaries of LLMs through this open-source 
effort.

Get access now at [LLM360 site](https://www.llm360.ai/)

# Instruction Tuning Training

**CrystalChat** is using the last **CrystalCoder** checkpoint of phase2 ([CrystalCoder_phase2_checkpoint_214387](https://huggingface.co/LLM360/CrystalCoder/tree/CrystalCoder_phase2_checkpoint_214387)) as the initialization checkpoint. We then finetune the model using the dataset mentioned below.

We also performed the same finetuning on the last **CrystalCoder** checkpoint of phase3 ([CrystalCoder_phase3_checkpoint_027728](https://huggingface.co/LLM360/CrystalCoder/tree/CrystalCoder_phase3_checkpoint_027728)). The phase2 and phase3 finetuning results are very similar, but phase2 finetuning exhibits slightly better performance on the English language benchmarks. We choose the phase2 finetuning result as the final model for **CrystalChat**.

# Instruction Tuning Data 

The fine-tuning data is a mix of publicly available language and code datasets, plus a orginally created dataset called **WebAlpaca** on HTML coding instructions.
The WebAlpaca dataset is created by us and is used as part of our instruction tuning training data. We will release the WebAlpaca dataset in a separate repository soon.

The summary of the fine-tuning data is as follows:

<!-- <center><img src="data_table.jpg" alt="Instruction Data"/></center> -->
| Subset      | #Tokens | Avg. #Q | Avg. Query Len | Avg. #R | Avg. Reply Len |
| ----------- | ----------- |----------- |----------- |----------- |----------- |
| [OASST1-guanaco](https://huggingface.co/datasets/openaccess-ai-collective/oasst1-guanaco-extended-sharegpt)      | 4,464,640       | 1.36 | 38.28 | 1.36 | 271.69 |
| [SlimOrca](https://huggingface.co/datasets/Open-Orca/SlimOrca)   |225,628,160        | 1.00 | 259.16	| 1.00	| 151.12 |
| [ShareGPT](https://huggingface.co/datasets/Aeala/ShareGPT_Vicuna_unfiltered)   | 112,914,432        | 3.28 | 94.53	| 3.64	| 365.81 | 
| [Evol-ShareGPT](https://huggingface.co/datasets/WizardLM/WizardLM_evol_instruct_V2_196k)   | 85,954,560        | 1.00	| 145.99 |	1.00	| 425.17 | 
| [ChatLogs](https://huggingface.co/datasets/winglian/chatlogs-en-cleaned)   | 29,337,600        | 3.39	| 95.58	| 3.24	| 191.42 |
| [CodeAlpaca](https://huggingface.co/datasets/lucasmccabe-lmi/CodeAlpaca-20k)   | 2,623,488        | 1.00	| 32.46	| 1.00	| 67.68 |
| [Rosetta Code](https://github.com/sahil280114/codealpaca/blob/master/data/rosetta_alpaca.json)   | 7,987,200        |  1.00 |	450.09	| 1.00	| 533.52 |
| [Evol-CodeAlpaca 1](https://huggingface.co/datasets/theblackcat102/evol-codealpaca-v1)   | 73,803,776        | 1.00	| 210.33 | 	1.00 | 	437.92 | 
| [Evol-CodeAlpaca 2](https://huggingface.co/datasets/nickrosh/Evol-Instruct-Code-80k-v1)   | 34,910,208        | 1.00	| 114.99 |	1.00 |	300.29 |
| WebAlpaca  | 43,673,600        | 1.00 |	96.29 |	1.00	| 746.52 | 
| [General Textbooks](https://huggingface.co/datasets/open-phi/textbooks)   | 85,590,016        | Not instruction data
| [Programming Books](https://huggingface.co/datasets/open-phi/programming_books_llama)   | 395,628,544        | Not instruction data
| Total | 1,102,516,224

For more details, check out the [data table](https://huggingface.co/LLM360/CrystalChat/blob/main/data_table.jpg).

# Instruction Format

We've added some new special tokens to the CrystalCoder tokenizer to support the instruction tuning.

List special tokens used in the instruction tuning:

```
bos: <s> 
eos: </s>
system_start: <|sys_start|>
system_end: <|sys_end|>
user_start: <|im_start|>
user_end: <|im_end|>
```

The instruction format is as follows:

```
<s> <|sys_start|> system prompt <|sys_end|> <|im_start|> first user utterance <|im_end|> first model response <|im_start|> next user utterance <|im_end|> next model response </s>
```

# Reproducing the Results

We will realize the training code and the training data soon. Our training code is based on [Megatron-LM](https://github.com/NVIDIA/Megatron-LM), with some modifications to support our training data format and Maximal Update Parametrization (μP).

# CrystalChat Performance

|           Model          | Trained Tokens | Avg. of Avg. | Language Avg. | Coding Avg. |  ARC  | HellaSwag | MMLU (5-shot) | GSM8K | Winogrande(5-shot) | TruthfulQA | HumanEval (pass@1) | MBPP (pass@1) |
|:------------------------:|:--------------:|:------------:|:-------------:|:-----------:|:-----:|:---------:|:-------------:|:-----:|:------------------:|:----------:|:------------------:|:-------------:|
| CrystalChat 7B           | 1.275T         | 44.96        | 53.29         | 36.62       | 51.71 | 76.12     | 53.22         | 28.05 | 70.64              | 47.29      | 34.12              | 39.11         |
| Mistral-7B-Instruct-v0.1 | -              | 44.34        | 54.86         | 30.62       | 58.05 | 75.71     | 55.56         | 32.00 | 74.27              | 55.90      | 29.27              | 31.96         |
| CodeLlama-7b-Instruct    | 2.5T           | 40.91        | 45.29         | 36.52       | 43.35 | 66.14     | 42.75         | 15.92 | 64.33              | 39.23      | 34.12              | 38.91         |
| Llama-2-7b-Chat          | 2T             | 34.11        | 52.86         | 15.35       | 53.07 | 78.39     | 48.42         | 18.88 | 73.09              | 45.30      | 13.26              | 17.43         |
| AmberChat 7B             | 1.25T          |     -        | 44.76         |     -       | 42.83 | 74.03     | 38.88         | 5.31  | 66.77              | 40.72      |     -              |       -       |



| Combined Language and Coding Ability           |
|------------------------------------------------|
<img src="CC-Compare.jpg" alt="arc" width="800"/>

| Performance on Standard Benchmarks             |
|------------------------------------------------|
<img src="cc-eval-std-benchmarks.png" alt="std-bench" width="800"/>

| Perforamnce on Language Benchmarks                      |
|---------------------------------------------------------|
<img src="cc-eval-lang-compare.png" alt="arc" width="800"/>

## Model Description

- **Model type:** Language model with the same architecture as LLaMA-7B
- **Language(s) (NLP):** English
- **License:** Apache 2.0
- **Resources for more information:**
  - [Training Code](https://github.com/LLM360/crystalcoder-train)
  - [Data Preparation](https://github.com/LLM360/crystalcoder-data-prep)
  - [Metrics](https://github.com/LLM360/Analysis360)
  - [Fully processed CrystalCoder pretraining data](https://huggingface.co/datasets/LLM360/CrystalCoderDatasets)

# Loading CrystalChat 

```python
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

device = "cuda:0" if torch.cuda.is_available() else "cpu"
tokenizer = AutoTokenizer.from_pretrained("LLM360/CrystalChat", trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained("LLM360/CrystalChat", trust_remote_code=True).to(device)

prompt = '<s> <|sys_start|> You are an AI assistant. You will be given a task. You must generate a detailed and long answer. <|sys_end|> <|im_start|> Write a python function that takes a list of integers and returns the squared sum of the list. <|im_end|>'


input_ids = tokenizer(prompt, return_tensors="pt").input_ids.to(device)
gen_tokens = model.generate(input_ids, do_sample=True, max_length=400)

print("-"*20 + "Output for model"  + 20 * '-')
print(tokenizer.batch_decode(gen_tokens)[0])
```

Response:
````
Here's a Python function named `squared_sum_list` that takes a list of integers as input and returns the squared sum of the list:

```python
def squared_sum_list(lst):
    return sum([num ** 2 for num in lst])
```

The function `squared_sum_list` uses a list comprehension to iterate over each number in the input list `lst` and calculate its square. Then, it uses the `sum` function to accumulate all the squared numbers in a single value - the squared sum of the list.

For example:

```python
numbers = [1, 2, 3, 4, 5]
print(squared_sum_list(numbers))  # Outputs: 55
```

In the above code, the list `[1, 2, 3, 4, 5]` is passed as an argument to the `squared_sum_list` function. The function calculates the sum of the squares of the elements in the list, which is `1 + 4 + 9 + 16 + 25 = 55`. The function then returns this result, which is printed to the console.</s>
````

# Evaluation

Coming Soon!

# Bias, Risks, and Limitations
CrystalChat has not been aligned to human preferences for safety within the RLHF phase or deployed with in-the-loop filtering of responses like ChatGPT, so the model can produce problematic outputs (especially when prompted to do so). The training data is known and made available [here](https://huggingface.co/datasets/LLM360/CrystalCoderDatasets). It primarily consists of SlimPajama, StarCoder, and WebCrawl dataset.

# Citation

**BibTeX:**

```bibtex
@misc{liu2023llm360,
      title={LLM360: Towards Fully Transparent Open-Source LLMs}, 
      author={Zhengzhong Liu and Aurick Qiao and Willie Neiswanger and Hongyi Wang and Bowen Tan and Tianhua Tao and Junbo Li and Yuqi Wang and Suqi Sun and Omkar Pangarkar and Richard Fan and Yi Gu and Victor Miller and Yonghao Zhuang and Guowei He and Haonan Li and Fajri Koto and Liping Tang and Nikhil Ranjan and Zhiqiang Shen and Xuguang Ren and Roberto Iriondo and Cun Mu and Zhiting Hu and Mark Schulze and Preslav Nakov and Tim Baldwin and Eric P. Xing},
      year={2023},
      eprint={2312.06550},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}
```