Update README.md

README.md

@@ -8,131 +8,21 @@ language:
 - en
 inference: false
 ---
-# Model Card for Mixtral-8x7B
-The Mixtral-8x7B Large Language Model (LLM) is a pretrained generative Sparse Mixture of Experts. The Mixtral-8x7B outperforms Llama 2 70B on most benchmarks we tested.
 
-For full details of this model please read our [release blog post](https://mistral.ai/news/mixtral-of-experts/).
+# Mixtral-8x7B (gpt-fast edition)
 
-## Warning
-This repo contains weights that are compatible with [vLLM](https://github.com/vllm-project/vllm) serving of the model as well as Hugging Face [transformers](https://github.com/huggingface/transformers) library. It is based on the original Mixtral [torrent release](magnet:?xt=urn:btih:5546272da9065eddeb6fcd7ffddeef5b75be79a7&dn=mixtral-8x7b-32kseqlen&tr=udp%3A%2F%http://2Fopentracker.i2p.rocks%3A6969%2Fannounce&tr=http%3A%2F%http://2Ftracker.openbittorrent.com%3A80%2Fannounce), but the file format and parameter names are different. Please note that model cannot (yet) be instantiated with HF.
+This repo holds quantized Mixtral-8x7B weights to be used in [gpt-fast](https://github.com/pytorch-labs/gpt-fast/).
 
-## Instruction format
+## Compatibility
 
-This format must be strictly respected, otherwise the model will generate sub-optimal outputs.
+Conversion to int4 was broken, so this repo only holds fp8 weights.
+Practically speaking this means your GPU(s) need to be Ada Lovelace or newer, and have enough VRAM to hold the model + KV cache + activations.
 
-The template used to build a prompt for the Instruct model is defined as follows:
-```
-<s> [INST] Instruction [/INST] Model answer</s> [INST] Follow-up instruction [/INST]
-```
-Note that `<s>` and `</s>` are special tokens for beginning of string (BOS) and end of string (EOS) while [INST] and [/INST] are regular strings.
+I'm hoping it can work on a pair of 4090s, which combined have 48 GiB (51.539607552 GB) of VRAM. Ignoring all overhead, this leaves
+~4.74 GB for KV-cache and activations, which should be enough (?).
 
-As reference, here is the pseudo-code used to tokenize instructions during fine-tuning:
-```python
-def tokenize(text):
-    return tok.encode(text, add_special_tokens=False)
+- [ ] TODO: Test on 2x4090 with TP=2
 
-[BOS_ID] + 
-tokenize("[INST]") + tokenize(USER_MESSAGE_1) + tokenize("[/INST]") +
-tokenize(BOT_MESSAGE_1) + [EOS_ID] +
-…
-tokenize("[INST]") + tokenize(USER_MESSAGE_N) + tokenize("[/INST]") +
-tokenize(BOT_MESSAGE_N) + [EOS_ID]
-```
+## Notes
 
-In the pseudo-code above, note that the `tokenize` method should not add a BOS or EOS token automatically, but should add a prefix space. 
-
-## Run the model
-
-```python
-from transformers import AutoModelForCausalLM, AutoTokenizer
-
-model_id = "mistralai/Mixtral-8x7B-Instruct-v0.1"
-tokenizer = AutoTokenizer.from_pretrained(model_id)
-
-model = AutoModelForCausalLM.from_pretrained(model_id)
-
-text = "Hello my name is"
-inputs = tokenizer(text, return_tensors="pt")
-
-outputs = model.generate(**inputs, max_new_tokens=20)
-print(tokenizer.decode(outputs[0], skip_special_tokens=True))
-```
-
-By default, transformers will load the model in full precision. Therefore you might be interested to further reduce down the memory requirements to run the model through the optimizations we offer in HF ecosystem:
-
-### In half-precision
-
-Note `float16` precision only works on GPU devices
-
-<details>
-<summary> Click to expand </summary>
-
-```diff
-+ import torch
-from transformers import AutoModelForCausalLM, AutoTokenizer
-
-model_id = "mistralai/Mixtral-8x7B-Instruct-v0.1"
-tokenizer = AutoTokenizer.from_pretrained(model_id)
-
-+ model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype=torch.float16).to(0)
-
-text = "Hello my name is"
-+ inputs = tokenizer(text, return_tensors="pt").to(0)
-
-outputs = model.generate(**inputs, max_new_tokens=20)
-print(tokenizer.decode(outputs[0], skip_special_tokens=True))
-```
-</details>
-
-### Lower precision using (8-bit & 4-bit) using `bitsandbytes`
-
-<details>
-<summary> Click to expand </summary>
-
-```diff
-+ import torch
-from transformers import AutoModelForCausalLM, AutoTokenizer
-
-model_id = "mistralai/Mixtral-8x7B-Instruct-v0.1"
-tokenizer = AutoTokenizer.from_pretrained(model_id)
-
-+ model = AutoModelForCausalLM.from_pretrained(model_id, load_in_4bit=True)
-
-text = "Hello my name is"
-+ inputs = tokenizer(text, return_tensors="pt").to(0)
-
-outputs = model.generate(**inputs, max_new_tokens=20)
-print(tokenizer.decode(outputs[0], skip_special_tokens=True))
-```
-</details>
-
-### Load the model with Flash Attention 2
-
-<details>
-<summary> Click to expand </summary>
-
-```diff
-+ import torch
-from transformers import AutoModelForCausalLM, AutoTokenizer
-
-model_id = "mistralai/Mixtral-8x7B-Instruct-v0.1"
-tokenizer = AutoTokenizer.from_pretrained(model_id)
-
-+ model = AutoModelForCausalLM.from_pretrained(model_id, use_flash_attention_2=True)
-
-text = "Hello my name is"
-+ inputs = tokenizer(text, return_tensors="pt").to(0)
-
-outputs = model.generate(**inputs, max_new_tokens=20)
-print(tokenizer.decode(outputs[0], skip_special_tokens=True))
-```
-</details>
-
-## Limitations
-
-The Mixtral-8x7B Instruct model is a quick demonstration that the base model can be easily fine-tuned to achieve compelling performance. 
-It does not have any moderation mechanisms. We're looking forward to engaging with the community on ways to
-make the model finely respect guardrails, allowing for deployment in environments requiring moderated outputs.
-
-# The Mistral AI Team
-Albert Jiang, Alexandre Sablayrolles, Arthur Mensch, Blanche Savary, Chris Bamford, Devendra Singh Chaplot, Diego de las Casas, Emma Bou Hanna, Florian Bressand, Gianna Lengyel, Guillaume Bour, Guillaume Lample, Lélio Renard Lavaud, Louis Ternon, Lucile Saulnier, Marie-Anne Lachaux, Pierre Stock, Teven Le Scao, Théophile Gervet, Thibaut Lavril, Thomas Wang, Timothée Lacroix, William El Sayed.
+Conversion was done with [(commit 7510a9d)](https://github.com/pytorch-labs/gpt-fast/commit/7510a9df7d23725ae46e9fca7d6ae8ee3a8f448e)