vanilj / tess-v2.5-qwen2-72b

我们创建了 Tess-v2.5，这是 Large Language Models (LLMs) 的 Tess 系列中的最新最先进的模型。Tess 是由 Migel Tissera 创建的旗舰级 LLM 系列，简称 Tesoro（意大利语中的宝藏）。Tess-v2.5 在推理能力、编码能力和数学能力方面取得了显著进展。它在 MMLU (Massive Multitask Language Understanding) 评估中是目前排名第一的开源权重模型。它在所有其他开源权重模型上的得分都高于 Qwen2-72B-Instruct，Llama3-70B-Instruct，Mixtral-8x22B-Instruct 和 DBRX-Instruct。进一步来说，当在 MMLU 评估时，Tess-v2.5 (Qwen2-72B) 模型的表现甚至超过了前沿的封闭模型 Gemini-1.0-Ultra，Gemini-1.5-Pro，Mistral-Large 和 Claude-3-Sonnet。

Tess-v2.5 (Qwen2-72B) 在新发布的 Qwen2-72B 基础上进行了微调，使用了包含 300K 样本的数据集 Tess-v2.5，这些样本覆盖了多个主题，包括商业和管理、市场营销、历史、社会科学、艺术、STEM 学科和计算机编程。此数据集是使用 Sensei 框架合成的，使用了多个前沿模型，例如 GPT-4-Turbo、Claude-Opus 和 Mistral-Large。

该模型的计算工作得到了 KindoAI 的大方赞助。

当在 AGIEval (Nous) 的子集上评估时，该模型与父模型 GPT-4-0314 的表现非常出色。

训练过程

Tess-v2.5 模型使用 Qwen2-72B 的基本权重启动。然后使用 Axolotl 作为训练框架，使用 Tess-v2.5 数据集进行微调。Tess 模型的大部分遵循共同微调方法：低学习率、低 epoch 数，并使用非常高质量且多样化的数据。该模型在 Microsoft Azure 上的 4xA100 VM 上进行了 4 天的微调。该模型尚未与 RLHF 或 DPO 一致。

作者认为，模型的能力似乎主要来自预训练过程。这是 Tess 模型每次微调的基础，而保留基础模型的熵对于作者是至关重要的。

评估结果

Tess-v2.5 模型是一个整体均衡良好的模型。有关该模型的所有评估结果都可以在评估文件夹中访问。

完整的评估比较表可以在此处访问：谷歌表格

MMLU (大量多任务语言理解)

AGIEval

运行推理的示例代码

请注意，此模型使用 ChatML 提示格式。

import torch, json
from transformers import AutoModelForCausalLM, AutoTokenizer
from stop_word import StopWordCriteria

model_path = "migtissera/Tess-v2.5-Qwen2-72B"
output_file_path = "/home/migel/conversations.jsonl"

model = AutoModelForCausalLM.from_pretrained(
    model_path,
    torch_dtype=torch.float16,
    device_map="auto",
    load_in_4bit=False,
    trust_remote_code=True,
)

tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True)

terminators = [
    tokenizer.convert_tokens_to_ids("<|im_end|>")
]

def generate_text(instruction):
    tokens = tokenizer.encode(instruction)
    tokens = torch.LongTensor(tokens).unsqueeze(0)
    tokens = tokens.to("cuda")

    instance = {
        "input_ids": tokens,
        "top_p": 1.0,
        "temperature": 0.75,
        "generate_len": 1024,
        "top_k": 50,
    }

    length = len(tokens[0])
    with torch.no_grad():
        rest = model.generate(
            input_ids=tokens,
            max_length=length + instance["generate_len"],
            use_cache=True,
            do_sample=True,
            top_p=instance["top_p"],
            temperature=instance["temperature"],
            top_k=instance["top_k"],
            num_return_sequences=1,
            pad_token_id=tokenizer.eos_token_id,
            eos_token_id=terminators,
        )
    output = rest[0][length:]
    string = tokenizer.decode(output, skip_special_tokens=True)
    return f"{string}"

conversation = f"""<|im_start|>system\nYou are Tesoro, a helful AI assitant. You always provide detailed answers without hesitation.<|im_end|>\n<|im_start|>user\n"""

while True:
    user_input = input("You: ")
    llm_prompt = f"{conversation}{user_input}<|im_end|>\n<|im_start|>assistant\n"
    answer = generate_text(llm_prompt)
    print(answer)
    conversation = f"{llm_prompt}{answer}\n"
    json_data = {"prompt": user_input, "answer": answer}

    with open(output_file_path, "a") as output_file:
        output_file.write(json.dumps(json_data) + "\n")

加入我的通用人工智能 Discord (NeuroLattice)

https://discord.gg/Hz6GrwGFKD

局限性 & 偏见

尽管该模型旨在提高准确性，但它偶尔可能会生成不准确或误导性的结果。

尽管对预训练数据进行了精心的优化，但仍然存在生成不当、有偏见或有冒犯性内容的可能性。

在必要时小心行事并进行交叉验证。这是一个未审核的模型。

Original model files can be found [here](https://hugging-face.cn/migtissera/Tess-v2.5-Qwen2-72B).

GGUF files can be found [here](https://hugging-face.cn/pabloce/Tess-v2.5-Qwen2-72B-gguf), Q4_K_M from [here](https://hugging-face.cn/bartowski/Tess-v2.5-Qwen2-72B-GGUF).

# Tess-v2.5 (Qwen2-72B)

![Tess-v2.5](https://hugging-face.cn/migtissera/Tess-v2.5-Qwen2-72B/resolve/main/Tess-v2.5.png)

We've created Tess-v2.5, the latest state-of-the-art model in the Tess series of Large Language Models (LLMs). Tess, short for Tesoro (<em>Treasure</em> in Italian), is the flagship LLM series created by Migel Tissera. Tess-v2.5 brings significant improvements in reasoning capabilities, coding capabilities and mathematics. It is currently the #1 ranked open weight model when evaluated on MMLU (Massive Multitask Language Understanding). It scores higher than all other open weight models including Qwen2-72B-Instruct, Llama3-70B-Instruct, Mixtral-8x22B-Instruct and DBRX-Instruct. Further, when evaluated on MMLU, Tess-v2.5 (Qwen2-72B) model outperforms even the frontier closed models Gemini-1.0-Ultra, Gemini-1.5-Pro, Mistral-Large and Claude-3-Sonnet.

Tess-v2.5 (Qwen2-72B) was fine-tuned over the newly released Qwen2-72B base, using the Tess-v2.5 dataset that contain 300K samples spanning multiple topics, including business and management, marketing, history, social sciences, arts, STEM subjects and computer programming. This dataset was synthetically generated using the [Sensei](https://github.com/migtissera/Sensei) framework, using multiple frontier models such as GPT-4-Turbo, Claude-Opus and Mistral-Large.

The compute for this model was generously sponsored by [KindoAI](https://kindo.ai).

When evaluated on a subset of AGIEval (Nous), this model compares very well with the godfather GPT-4-0314 model as well.

# Training Process

Tess-v2.5 model was initiated with the base weights of Qwen2-72B. It was then fine-tuned with the Tess-v2.5 dataset, using Axolotl as the training framework. Most of Tess models follow a common fine-tuning methodology: low learning rates, low number of epochs, and uses very high quality and diverse data. This model was fine-tuned on a 4xA100 VM on Microsoft Azure for 4 days. The model has not been aligned with RLHF or DPO.

The author believes that model's capabilities seem to come primariliy from the pre-training process. This is the foundation for every fine-tune of Tess models, and preserving the entropy of the base models is of paramount to the author.

# Evaluation Results

Tess-v2.5 model is an overall well balanced model. All evals pertaining to this model can be accessed in the [Evals](https://hugging-face.cn/migtissera/Tess-v2.5-Qwen2-72B/tree/main/Evals) folder.

Complete evaluation comparison tables can be accessed here: [Google Spreadsheet](https://docs.google.com/spreadsheets/d/1k0BIKux_DpuoTPwFCTMBzczw17kbpxofigHF_0w2LGw/edit?usp=sharing)

## MMLU (Massive Multitask Language Understanding)
![MMLU_open](https://hugging-face.cn/migtissera/Tess-v2.5-Qwen2-72B/resolve/main/Figures/mmlu_open_models.png)

![MMLU_closed](https://hugging-face.cn/migtissera/Tess-v2.5-Qwen2-72B/resolve/main/Figures/mmlu_closed_models.png)

## AGIEval
![AGIEval](https://hugging-face.cn/migtissera/Tess-v2.5-Qwen2-72B/resolve/main/Figures/AGIEval.png)

# Sample code to run inference

Note that this model uses ChatML prompt format.

```python
import torch, json
from transformers import AutoModelForCausalLM, AutoTokenizer
from stop_word import StopWordCriteria

model_path = "migtissera/Tess-v2.5-Qwen2-72B"
output_file_path = "/home/migel/conversations.jsonl"

model = AutoModelForCausalLM.from_pretrained(
    model_path,
    torch_dtype=torch.float16,
    device_map="auto",
    load_in_4bit=False,
    trust_remote_code=True,
)

tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True)

terminators = [
    tokenizer.convert_tokens_to_ids("<|im_end|>")
]

def generate_text(instruction):
    tokens = tokenizer.encode(instruction)
    tokens = torch.LongTensor(tokens).unsqueeze(0)
    tokens = tokens.to("cuda")

instance = {
        "input_ids": tokens,
        "top_p": 1.0,
        "temperature": 0.75,
        "generate_len": 1024,
        "top_k": 50,
    }

length = len(tokens[0])
    with torch.no_grad():
        rest = model.generate(
            input_ids=tokens,
            max_length=length + instance["generate_len"],
            use_cache=True,
            do_sample=True,
            top_p=instance["top_p"],
            temperature=instance["temperature"],
            top_k=instance["top_k"],
            num_return_sequences=1,
            pad_token_id=tokenizer.eos_token_id,
            eos_token_id=terminators,
        )
    output = rest[0][length:]
    string = tokenizer.decode(output, skip_special_tokens=True)
    return f"{string}"

while True:
    user_input = input("You: ")
    llm_prompt = f"{conversation}{user_input}<|im_end|>\n<|im_start|>assistant\n"
    answer = generate_text(llm_prompt)
    print(answer)
    conversation = f"{llm_prompt}{answer}\n"
    json_data = {"prompt": user_input, "answer": answer}

with open(output_file_path, "a") as output_file:
        output_file.write(json.dumps(json_data) + "\n")
```

# Join My General AI Discord (NeuroLattice):
https://discord.gg/Hz6GrwGFKD

# Limitations & Biases:

While this model aims for accuracy, it can occasionally produce inaccurate or misleading results.

Despite diligent efforts in refining the pretraining data, there remains a possibility for the generation of inappropriate, biased, or offensive content.

Exercise caution and cross-check information when necessary. This is an uncensored model.

Paste, drop or click to upload images (.png, .jpeg, .jpg, .svg, .gif)