Update README.md

README.md

@@ -15,13 +15,13 @@ language:
 <em>[Paper][Code][🤗] (would be released soon)</em>
 </p>
 
-Infinity-Instruct-6M-0718-Mistral-7B is an opensource supervised instruction tuning model without reinforcement learning from human feedback (RLHF). This model is just finetuned on [Infinity-Instruct-6M and Infinity-Instruct-0718](https://huggingface.co/datasets/BAAI/Infinity-Instruct) and showing favorable results on AlpacaEval 2.0 compared to Mixtral 8x22B v0.1, Gemini Pro, and GPT-4.
+Infinity-Instruct-7M-0729-Mistral-7B is an opensource supervised instruction tuning model without reinforcement learning from human feedback (RLHF). This model is just finetuned on [Infinity-Instruct-7M and Infinity-Instruct-0729](https://huggingface.co/datasets/BAAI/Infinity-Instruct) and showing favorable results on AlpacaEval 2.0 compared to Mixtral 8x22B v0.1, Gemini Pro, and GPT-4.
 
 ## **News**
 
-- 🔥🔥🔥[2024/07/18] We release the model weights of [InfInstruct-Mistral-7B 0718](https://huggingface.co/BAAI/Infinity-Instruct-3M-0718-mistral-7B), [InfInstruct-Llama3-70B 0718](https://huggingface.co/BAAI/Infinity-Instruct-3M-0718-Llama3-70B).
+- 🔥🔥🔥[2024/08/02] We release the model weights of [InfInstruct-Llama3.1-70B 0729](https://huggingface.co/BAAI/Infinity-Instruct-7M-0729-Llama3_1-70B), [InfInstruct-Llama3.1-8B 0729](https://huggingface.co/BAAI/Infinity-Instruct-7M-0729-Llama3_1-70B), [InfInstruct-Mistral-7B 0729](https://huggingface.co/BAAI/Infinity-Instruct-7M-0729-Mistral-7B).
 
-- 🔥🔥🔥[2024/07/18] We release the foundational dataset [Infinity-Instruct-6M](https://huggingface.co/datasets/BAAI/Infinity-Instruct) the chat dataset [Infinity-Instruct-0718](https://huggingface.co/datasets/BAAI/Infinity-Instruct).
+- 🔥🔥🔥[2024/08/02] We release the 7M foundational dataset [Infinity-Instruct-7M](https://huggingface.co/datasets/BAAI/Infinity-Instruct).
 
 - 🔥🔥🔥[2024/07/09] We release the model weights of [InfInstruct-Mistral-7B 0625](https://huggingface.co/BAAI/Infinity-Instruct-3M-0625-Mistral-7B), [InfInstruct-Qwen2-7B 0625](https://huggingface.co/BAAI/Infinity-Instruct-3M-0625-Qwen2-7B), [InfInstruct-Llama3-8B 0625](https://huggingface.co/BAAI/Infinity-Instruct-3M-0625-Llama3-8B), [InfInstruct-Llama3-70B 0625](https://huggingface.co/BAAI/Infinity-Instruct-3M-0625-Llama3-70B), and [InfInstruct-Yi-1.5-9B 0625](https://huggingface.co/BAAI/Infinity-Instruct-3M-0625-Yi-1.5-9B).
 
@@ -39,7 +39,7 @@ Infinity-Instruct-6M-0718-Mistral-7B is an opensource supervised instruction tun
 <img src="fig/trainingflow.png">
 </p>
 
-Infinity-Instruct-6M-0718-Mistral-7B is tuned on Million-level instruction dataset [Infinity-Instruct](https://huggingface.co/datasets/BAAI/Infinity-Instruct). First, we apply the foundational dataset Infinity-Instruct-6M to improve the foundational ability (math & code) of Mistral-7B-v0.1, and get the foundational instruct model Infinity-Instruct-6M-Mistral-7B. Then we finetune the Infinity-Instruct-6M-Mistral-7B to get the stronger chat model Infinity-Instruct-6M-0718-Mistral-7B. Here is the training hyperparamers. 
+Infinity-Instruct-7M-0729-Mistral-7B is tuned on Million-level instruction dataset [Infinity-Instruct](https://huggingface.co/datasets/BAAI/Infinity-Instruct). First, we apply the foundational dataset Infinity-Instruct-7M to improve the foundational ability (math & code) of Mistral-7B-v0.1, and get the foundational instruct model Infinity-Instruct-7M-Mistral-7B. Then we finetune the Infinity-Instruct-7M-Mistral-7B to get the stronger chat model Infinity-Instruct-7M-0729-Mistral-7B. Here is the training hyperparamers. 
 
 ```bash
 epoch: 3
@@ -58,27 +58,23 @@ Thanks to [FlagScale](https://github.com/FlagOpen/FlagScale), we could concatena
 
 ## **Benchmark**
 
-|            **Model**            | **MT-Bench** | **AlpacaEval2.0** |
-|:-------------------------------:|:------------:|:-----------------:|
-| OpenHermes-2.5-Mistral-7B*      |      7.5     |        16.2       |
-| Mistral-7B-Instruct-v0.2        |      7.6     |        17.1       |
-| Llama-3-8B-Instruct             |      8.1     |        22.9       |
-| GPT-4-0314                      |      9.0     |        35.3       |
-| Mixtral 8x7B v0.1               |      8.3     |        23.7       |
-| Mixtral 8x22B v0.1              |      --      |        30.9       |
-| Gemini Pro                      |      --      |        24.4       |
-| InfInstruct-3M-Mistral-7B*      |      7.6     |        16.2       |
-| InfInstruct-3M-0613-Mistral-7B* |      8.1     |        25.5       |
-| InfInstruct-3M-0625-Mistral-7B* |      8.1     |         31.42     |
-| InfInstruct-3M-0718-Mistral-7B* |      8.1     |         **40.0**  |
+|           **Model**          | **MT-Bench** | **AlpacaEval2.0** | **Arena-hard** |
+|:----------------------------:|:------------:|:-----------------:|:-----------------:|
+| GPT-4-0314                   |      9.0     |        35.3       |        50.0       |
+| GPT-4-0613                   |      9.2     |        30.2       |        37.9       |
+| GPT-4-1106                   |      9.3     |        30.2       |        --       |
+| Gemini Pro                   |      --      |        24.4       |        17.8       |
+| Mixtral 8x7B v0.1            |      8.3     |        23.7       |        23.4       |
+| Mistral-7B-Instruct-v0.2     |      7.6     |        17.1       |        --       | 
+| InfInstruct-3M-0613-Mistral-7B  |      8.1     |        25.5       |        --       |
+| InfInstruct-3M-0625-Mistral-7B  |      8.1     |        31.4       |        --       |
+| **InfInstruct-3M-0629-Mistral-7B**  |      **8.1**     |        **40.0**       |        **26.9**       |
 
 *denote the model is finetuned without reinforcement learning from human feedback (RLHF). 
 
-We evaluate Infinity-Instruct-6M-0718-Mistral-7B on the two most popular instructions following benchmarks. Mt-Bench is a set of challenging multi-turn questions including code, math and routine dialogue. AlpacaEval2.0 is based on AlpacaFarm evaluation set. Both of these two benchmarks use GPT-4 to judge the model answer. AlpacaEval2.0 displays a high agreement rate with human-annotated benchmark, Chatbot Arena. The result shows that InfInstruct-6M-0718-Mistral-7B achieved 40.0 in AlpacaEval2.0, which is higher than the 35.5 of GPT4-0314 although it does not yet use RLHF. InfInstruct-6M-0718-Mistral-7B also achieves 8.1 in MT-Bench, which is comparable to the state-of-the-art billion-parameter LLM such as Llama-3-8B-Instruct and Mistral-7B-Instruct-v0.2.
-
 ## **How to use**
 
-Infinity-Instruct-6M-0718-Mistral-7B adopt the same chat template of [OpenHermes-2.5-Mistral-7B](https://huggingface.co/teknium/OpenHermes-2.5-Mistral-7B):
+Infinity-Instruct-7M-0729-Mistral-7B adopt the same chat template of [OpenHermes-2.5-Mistral-7B](https://huggingface.co/teknium/OpenHermes-2.5-Mistral-7B):
 
 ```bash
 <|im_start|>system
@@ -98,11 +94,11 @@ from transformers import AutoModelForCausalLM, AutoTokenizer, LogitsProcessorLis
 import torch
 device = "cuda" # the device to load the model onto
 
-model = AutoModelForCausalLM.from_pretrained("BAAI/Infinity-Instruct-6M-0718-Mistral-7B",
+model = AutoModelForCausalLM.from_pretrained("BAAI/Infinity-Instruct-7M-0729-Mistral-7B",
     torch_dtype=torch.bfloat16,
     device_map="auto"
 )
-tokenizer = AutoTokenizer.from_pretrained("BAAI/Infinity-Instruct-6M-0718-Mistral-7BB")
+tokenizer = AutoTokenizer.from_pretrained("BAAI/Infinity-Instruct-7M-0729-Mistral-7BB")
 
 # This template is copied from OpenHermers-mistral-2.5 (https://huggingface.co/teknium/OpenHermes-2.5-Mistral-7B)
 prompt = "Give me a short introduction to large language model."