--- language: - en license: llama2 tags: - upstage - llama-2 - instruct - instruction model_name: Llama 2 70B Instruct v2 inference: false model_creator: Upstage model_link: https://huggingface.co/upstage/Llama-2-70b-instruct-v2 model_type: llama pipeline_tag: text-generation quantized_by: TheBloke base_model: upstage/Llama-2-70b-instruct-v2 ---
TheBlokeAI

Chat & support: TheBloke's Discord server

Want to contribute? TheBloke's Patreon page

TheBloke's LLM work is generously supported by a grant from andreessen horowitz (a16z)

# Llama 2 70B Instruct v2 - GGML - Model creator: [Upstage](https://huggingface.co/Upstage) - Original model: [Llama 2 70B Instruct v2](https://huggingface.co/upstage/Llama-2-70b-instruct-v2) ## Description This repo contains GGML format model files for [Upstage's Llama 2 70B Instruct v2](https://huggingface.co/upstage/Llama-2-70b-instruct-v2). ### Important note regarding GGML files. The GGML format has now been superseded by GGUF. As of August 21st 2023, [llama.cpp](https://github.com/ggerganov/llama.cpp) no longer supports GGML models. Third party clients and libraries are expected to still support it for a time, but many may also drop support. Please use the GGUF models instead. ### About GGML GPU acceleration is now available for Llama 2 70B GGML files, with both CUDA (NVidia) and Metal (macOS). The following clients/libraries are known to work with these files, including with GPU acceleration: * [llama.cpp](https://github.com/ggerganov/llama.cpp), commit `e76d630` and later. * [text-generation-webui](https://github.com/oobabooga/text-generation-webui), the most widely used web UI. * [KoboldCpp](https://github.com/LostRuins/koboldcpp), version 1.37 and later. A powerful GGML web UI, especially good for story telling. * [LM Studio](https://lmstudio.ai/), a fully featured local GUI with GPU acceleration for both Windows and macOS. Use 0.1.11 or later for macOS GPU acceleration with 70B models. * [llama-cpp-python](https://github.com/abetlen/llama-cpp-python), version 0.1.77 and later. A Python library with LangChain support, and OpenAI-compatible API server. * [ctransformers](https://github.com/marella/ctransformers), version 0.2.15 and later. A Python library with LangChain support, and OpenAI-compatible API server. ## Repositories available * [GPTQ models for GPU inference, with multiple quantisation parameter options.](https://huggingface.co/TheBloke/Upstage-Llama-2-70B-instruct-v2-GPTQ) * [2, 3, 4, 5, 6 and 8-bit GGUF models for CPU+GPU inference](https://huggingface.co/TheBloke/Upstage-Llama-2-70B-instruct-v2-GGUF) * [2, 3, 4, 5, 6 and 8-bit GGML models for CPU+GPU inference (deprecated)](https://huggingface.co/TheBloke/Upstage-Llama-2-70B-instruct-v2-GGML) * [Upstage's original unquantised fp16 model in pytorch format, for GPU inference and for further conversions](https://huggingface.co/upstage/Llama-2-70b-instruct-v2) ## Prompt template: Orca-Hashes ``` ### System: {system_message} ### User: {prompt} ### Assistant: ``` ## Compatibility ### Works with llama.cpp [commit `e76d630`](https://github.com/ggerganov/llama.cpp/commit/e76d630df17e235e6b9ef416c45996765d2e36fb) until August 21st, 2023 Will not work with `llama.cpp` after commit [dadbed99e65252d79f81101a392d0d6497b86caa](https://github.com/ggerganov/llama.cpp/commit/dadbed99e65252d79f81101a392d0d6497b86caa). For compatibility with latest llama.cpp, please use GGUF files instead. Or one of the other tools and libraries listed above. To use in llama.cpp, you must add `-gqa 8` argument. For other UIs and libraries, please check the docs. ## Explanation of the new k-quant methods
Click to see details The new methods available are: * GGML_TYPE_Q2_K - "type-1" 2-bit quantization in super-blocks containing 16 blocks, each block having 16 weight. Block scales and mins are quantized with 4 bits. This ends up effectively using 2.5625 bits per weight (bpw) * GGML_TYPE_Q3_K - "type-0" 3-bit quantization in super-blocks containing 16 blocks, each block having 16 weights. Scales are quantized with 6 bits. This end up using 3.4375 bpw. * GGML_TYPE_Q4_K - "type-1" 4-bit quantization in super-blocks containing 8 blocks, each block having 32 weights. Scales and mins are quantized with 6 bits. This ends up using 4.5 bpw. * GGML_TYPE_Q5_K - "type-1" 5-bit quantization. Same super-block structure as GGML_TYPE_Q4_K resulting in 5.5 bpw * GGML_TYPE_Q6_K - "type-0" 6-bit quantization. Super-blocks with 16 blocks, each block having 16 weights. Scales are quantized with 8 bits. This ends up using 6.5625 bpw * GGML_TYPE_Q8_K - "type-0" 8-bit quantization. Only used for quantizing intermediate results. The difference to the existing Q8_0 is that the block size is 256. All 2-6 bit dot products are implemented for this quantization type. Refer to the Provided Files table below to see what files use which methods, and how.
## Provided files | Name | Quant method | Bits | Size | Max RAM required | Use case | | ---- | ---- | ---- | ---- | ---- | ----- | | [upstage-llama-2-70b-instruct-v2.ggmlv3.q2_K.bin](https://huggingface.co/TheBloke/Upstage-Llama-2-70B-instruct-v2-GGML/blob/main/upstage-llama-2-70b-instruct-v2.ggmlv3.q2_K.bin) | q2_K | 2 | 28.59 GB| 31.09 GB | New k-quant method. Uses GGML_TYPE_Q4_K for the attention.vw and feed_forward.w2 tensors, GGML_TYPE_Q2_K for the other tensors. | | [upstage-llama-2-70b-instruct-v2.ggmlv3.q3_K_S.bin](https://huggingface.co/TheBloke/Upstage-Llama-2-70B-instruct-v2-GGML/blob/main/upstage-llama-2-70b-instruct-v2.ggmlv3.q3_K_S.bin) | q3_K_S | 3 | 29.75 GB| 32.25 GB | New k-quant method. Uses GGML_TYPE_Q3_K for all tensors | | [upstage-llama-2-70b-instruct-v2.ggmlv3.q3_K_M.bin](https://huggingface.co/TheBloke/Upstage-Llama-2-70B-instruct-v2-GGML/blob/main/upstage-llama-2-70b-instruct-v2.ggmlv3.q3_K_M.bin) | q3_K_M | 3 | 33.04 GB| 35.54 GB | New k-quant method. Uses GGML_TYPE_Q4_K for the attention.wv, attention.wo, and feed_forward.w2 tensors, else GGML_TYPE_Q3_K | | [upstage-llama-2-70b-instruct-v2.ggmlv3.q3_K_L.bin](https://huggingface.co/TheBloke/Upstage-Llama-2-70B-instruct-v2-GGML/blob/main/upstage-llama-2-70b-instruct-v2.ggmlv3.q3_K_L.bin) | q3_K_L | 3 | 36.15 GB| 38.65 GB | New k-quant method. Uses GGML_TYPE_Q5_K for the attention.wv, attention.wo, and feed_forward.w2 tensors, else GGML_TYPE_Q3_K | | [upstage-llama-2-70b-instruct-v2.ggmlv3.q4_0.bin](https://huggingface.co/TheBloke/Upstage-Llama-2-70B-instruct-v2-GGML/blob/main/upstage-llama-2-70b-instruct-v2.ggmlv3.q4_0.bin) | q4_0 | 4 | 38.87 GB| 41.37 GB | Original quant method, 4-bit. | | [upstage-llama-2-70b-instruct-v2.ggmlv3.q4_K_S.bin](https://huggingface.co/TheBloke/Upstage-Llama-2-70B-instruct-v2-GGML/blob/main/upstage-llama-2-70b-instruct-v2.ggmlv3.q4_K_S.bin) | q4_K_S | 4 | 38.87 GB| 41.37 GB | New k-quant method. Uses GGML_TYPE_Q4_K for all tensors | | [upstage-llama-2-70b-instruct-v2.ggmlv3.q4_K_M.bin](https://huggingface.co/TheBloke/Upstage-Llama-2-70B-instruct-v2-GGML/blob/main/upstage-llama-2-70b-instruct-v2.ggmlv3.q4_K_M.bin) | q4_K_M | 4 | 41.38 GB| 43.88 GB | New k-quant method. Uses GGML_TYPE_Q6_K for half of the attention.wv and feed_forward.w2 tensors, else GGML_TYPE_Q4_K | | [upstage-llama-2-70b-instruct-v2.ggmlv3.q4_1.bin](https://huggingface.co/TheBloke/Upstage-Llama-2-70B-instruct-v2-GGML/blob/main/upstage-llama-2-70b-instruct-v2.ggmlv3.q4_1.bin) | q4_1 | 4 | 43.17 GB| 45.67 GB | Original quant method, 4-bit. Higher accuracy than q4_0 but not as high as q5_0. However has quicker inference than q5 models. | | [upstage-llama-2-70b-instruct-v2.ggmlv3.q5_0.bin](https://huggingface.co/TheBloke/Upstage-Llama-2-70B-instruct-v2-GGML/blob/main/upstage-llama-2-70b-instruct-v2.ggmlv3.q5_0.bin) | q5_0 | 5 | 47.46 GB| 49.96 GB | Original quant method, 5-bit. Higher accuracy, higher resource usage and slower inference. | | [upstage-llama-2-70b-instruct-v2.ggmlv3.q5_K_S.bin](https://huggingface.co/TheBloke/Upstage-Llama-2-70B-instruct-v2-GGML/blob/main/upstage-llama-2-70b-instruct-v2.ggmlv3.q5_K_S.bin) | q5_K_S | 5 | 47.46 GB| 49.96 GB | New k-quant method. Uses GGML_TYPE_Q5_K for all tensors | | [upstage-llama-2-70b-instruct-v2.ggmlv3.q5_K_M.bin](https://huggingface.co/TheBloke/Upstage-Llama-2-70B-instruct-v2-GGML/blob/main/upstage-llama-2-70b-instruct-v2.ggmlv3.q5_K_M.bin) | q5_K_M | 5 | 48.75 GB| 51.25 GB | New k-quant method. Uses GGML_TYPE_Q6_K for half of the attention.wv and feed_forward.w2 tensors, else GGML_TYPE_Q5_K | **Note**: the above RAM figures assume no GPU offloading. If layers are offloaded to the GPU, this will reduce RAM usage and use VRAM instead. ## How to run in `llama.cpp` Make sure you are using `llama.cpp` from commit [dadbed99e65252d79f81101a392d0d6497b86caa](https://github.com/ggerganov/llama.cpp/commit/dadbed99e65252d79f81101a392d0d6497b86caa) or earlier. For compatibility with latest llama.cpp, please use GGUF files instead. I use the following command line; adjust for your tastes and needs: ``` ./main -t 10 -ngl 40 -gqa 8 -m upstage-llama-2-70b-instruct-v2.ggmlv3.q4_K_M.bin --color -c 4096 --temp 0.7 --repeat_penalty 1.1 -n -1 -p "### System:\n{system_message}\n\n### User:\n{prompt}\n\n### Assistant:" ``` Change `-t 10` to the number of physical CPU cores you have. For example if your system has 8 cores/16 threads, use `-t 8`. If you are fully offloading the model to GPU, use `-t 1` Change `-ngl 40` to the number of GPU layers you have VRAM for. Use `-ngl 100` to offload all layers to VRAM - if you have a 48GB card, or 2 x 24GB, or similar. Otherwise you can partially offload as many as you have VRAM for, on one or more GPUs. If you want to have a chat-style conversation, replace the `-p ` argument with `-i -ins` Remember the `-gqa 8` argument, required for Llama 70B models. Change `-c 4096` to the desired sequence length for this model. For models that use RoPE, add `--rope-freq-base 10000 --rope-freq-scale 0.5` for doubled context, or `--rope-freq-base 10000 --rope-freq-scale 0.25` for 4x context. For other parameters and how to use them, please refer to [the llama.cpp documentation](https://github.com/ggerganov/llama.cpp/blob/master/examples/main/README.md) ## How to run in `text-generation-webui` Further instructions here: [text-generation-webui/docs/llama.cpp-models.md](https://github.com/oobabooga/text-generation-webui/blob/main/docs/llama.cpp-models.md). ## Discord For further support, and discussions on these models and AI in general, join us at: [TheBloke AI's Discord server](https://discord.gg/theblokeai) ## Thanks, and how to contribute. Thanks to the [chirper.ai](https://chirper.ai) team! I've had a lot of people ask if they can contribute. I enjoy providing models and helping people, and would love to be able to spend even more time doing it, as well as expanding into new projects like fine tuning/training. If you're able and willing to contribute it will be most gratefully received and will help me to keep providing more models, and to start work on new AI projects. Donaters will get priority support on any and all AI/LLM/model questions and requests, access to a private Discord room, plus other benefits. * Patreon: https://patreon.com/TheBlokeAI * Ko-Fi: https://ko-fi.com/TheBlokeAI **Special thanks to**: Aemon Algiz. **Patreon special mentions**: Russ Johnson, J, alfie_i, Alex, NimbleBox.ai, Chadd, Mandus, Nikolai Manek, Ken Nordquist, ya boyyy, Illia Dulskyi, Viktor Bowallius, vamX, Iucharbius, zynix, Magnesian, Clay Pascal, Pierre Kircher, Enrico Ros, Tony Hughes, Elle, Andrey, knownsqashed, Deep Realms, Jerry Meng, Lone Striker, Derek Yates, Pyrater, Mesiah Bishop, James Bentley, Femi Adebogun, Brandon Frisco, SuperWojo, Alps Aficionado, Michael Dempsey, Vitor Caleffi, Will Dee, Edmond Seymore, usrbinkat, LangChain4j, Kacper Wikieł, Luke Pendergrass, John Detwiler, theTransient, Nathan LeClaire, Tiffany J. Kim, biorpg, Eugene Pentland, Stanislav Ovsiannikov, Fred von Graf, terasurfer, Kalila, Dan Guido, Nitin Borwankar, 阿明, Ai Maven, John Villwock, Gabriel Puliatti, Stephen Murray, Asp the Wyvern, danny, Chris Smitley, ReadyPlayerEmma, S_X, Daniel P. Andersen, Olakabola, Jeffrey Morgan, Imad Khwaja, Caitlyn Gatomon, webtim, Alicia Loh, Trenton Dambrowitz, Swaroop Kallakuri, Erik Bjäreholt, Leonard Tan, Spiking Neurons AB, Luke @flexchar, Ajan Kanaga, Thomas Belote, Deo Leter, RoA, Willem Michiel, transmissions 11, subjectnull, Matthew Berman, Joseph William Delisle, David Ziegler, Michael Davis, Johann-Peter Hartmann, Talal Aujan, senxiiz, Artur Olbinski, Rainer Wilmers, Spencer Kim, Fen Risland, Cap'n Zoog, Rishabh Srivastava, Michael Levine, Geoffrey Montalvo, Sean Connelly, Alexandros Triantafyllidis, Pieter, Gabriel Tamborski, Sam, Subspace Studios, Junyu Yang, Pedro Madruga, Vadim, Cory Kujawski, K, Raven Klaugh, Randy H, Mano Prime, Sebastain Graf, Space Cruiser Thank you to all my generous patrons and donaters! And thank you again to a16z for their generous grant. # Original model card: Upstage's Llama 2 70B Instruct v2 # SOLAR-0-70b-16bit model card The model name has been changed from LLaMa-2-70b-instruct-v2 to SOLAR-0-70b-16bit ## Model Details * **Developed by**: [Upstage](https://en.upstage.ai) * **Backbone Model**: [LLaMA-2](https://github.com/facebookresearch/llama/tree/main) * **Language(s)**: English * **Library**: [HuggingFace Transformers](https://github.com/huggingface/transformers) * **License**: Fine-tuned checkpoints is licensed under the Non-Commercial Creative Commons license ([CC BY-NC-4.0](https://creativecommons.org/licenses/by-nc/4.0/)) * **Where to send comments**: Instructions on how to provide feedback or comments on a model can be found by opening an issue in the [Hugging Face community's model repository](https://huggingface.co/upstage/Llama-2-70b-instruct-v2/discussions) * **Contact**: For questions and comments about the model, please email [contact@upstage.ai](mailto:contact@upstage.ai) ## Dataset Details ### Used Datasets - Orca-style dataset - Alpaca-style dataset - No other dataset was used except for the dataset mentioned above - No benchmark test set or the training set are used ### Prompt Template ``` ### System: {System} ### User: {User} ### Assistant: {Assistant} ``` ## Usage - The followings are tested on A100 80GB - Our model can handle up to 10k+ input tokens, thanks to the `rope_scaling` option ```python import torch from transformers import AutoModelForCausalLM, AutoTokenizer, TextStreamer tokenizer = AutoTokenizer.from_pretrained("upstage/Llama-2-70b-instruct-v2") model = AutoModelForCausalLM.from_pretrained( "upstage/Llama-2-70b-instruct-v2", device_map="auto", torch_dtype=torch.float16, load_in_8bit=True, rope_scaling={"type": "dynamic", "factor": 2} # allows handling of longer inputs ) prompt = "### User:\nThomas is healthy, but he has to go to the hospital. What could be the reasons?\n\n### Assistant:\n" inputs = tokenizer(prompt, return_tensors="pt").to(model.device) del inputs["token_type_ids"] streamer = TextStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True) output = model.generate(**inputs, streamer=streamer, use_cache=True, max_new_tokens=float('inf')) output_text = tokenizer.decode(output[0], skip_special_tokens=True) ``` ## Hardware and Software * **Hardware**: We utilized an A100x8 * 4 for training our model * **Training Factors**: We fine-tuned this model using a combination of the [DeepSpeed library](https://github.com/microsoft/DeepSpeed) and the [HuggingFace Trainer](https://huggingface.co/docs/transformers/main_classes/trainer) / [HuggingFace Accelerate](https://huggingface.co/docs/accelerate/index) ## Evaluation Results ### Overview - We conducted a performance evaluation following the tasks being evaluated on the [Open LLM Leaderboard](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard). We evaluated our model on four benchmark datasets, which include `ARC-Challenge`, `HellaSwag`, `MMLU`, and `TruthfulQA` We used the [lm-evaluation-harness repository](https://github.com/EleutherAI/lm-evaluation-harness), specifically commit [b281b0921b636bc36ad05c0b0b0763bd6dd43463](https://github.com/EleutherAI/lm-evaluation-harness/tree/b281b0921b636bc36ad05c0b0b0763bd6dd43463). - We used [MT-bench](https://github.com/lm-sys/FastChat/tree/main/fastchat/llm_judge), a set of challenging multi-turn open-ended questions, to evaluate the models ### Main Results | Model | H4(Avg) | ARC | HellaSwag | MMLU | TruthfulQA | | MT_Bench | |--------------------------------------------------------------------|----------|----------|----------|------|----------|-|-------------| | **[Llama-2-70b-instruct-v2](https://huggingface.co/upstage/Llama-2-70b-instruct-v2)**(***Ours***, ***Open LLM Leaderboard***) | **73** | **71.1** | **87.9** | **70.6** | **62.2** | | **7.44063** | | [Llama-2-70b-instruct](https://huggingface.co/upstage/Llama-2-70b-instruct) (Ours, Open LLM Leaderboard) | 72.3 | 70.9 | 87.5 | 69.8 | 61 | | 7.24375 | | [llama-65b-instruct](https://huggingface.co/upstage/llama-65b-instruct) (Ours, Open LLM Leaderboard) | 69.4 | 67.6 | 86.5 | 64.9 | 58.8 | | | | Llama-2-70b-hf | 67.3 | 67.3 | 87.3 | 69.8 | 44.9 | | | | [llama-30b-instruct-2048](https://huggingface.co/upstage/llama-30b-instruct-2048) (Ours, Open LLM Leaderboard) | 67.0 | 64.9 | 84.9 | 61.9 | 56.3 | | | | [llama-30b-instruct](https://huggingface.co/upstage/llama-30b-instruct) (Ours, Open LLM Leaderboard) | 65.2 | 62.5 | 86.2 | 59.4 | 52.8 | | | | llama-65b | 64.2 | 63.5 | 86.1 | 63.9 | 43.4 | | | | falcon-40b-instruct | 63.4 | 61.6 | 84.3 | 55.4 | 52.5 | | | ### Scripts for H4 Score Reproduction - Prepare evaluation environments: ``` # clone the repository git clone https://github.com/EleutherAI/lm-evaluation-harness.git # check out the specific commit git checkout b281b0921b636bc36ad05c0b0b0763bd6dd43463 # change to the repository directory cd lm-evaluation-harness ``` ## Contact Us ### About Upstage - [Upstage](https://en.upstage.ai) is a company specialized in Large Language Models (LLMs) and AI. We will help you build private LLMs and related applications. If you have a dataset to build domain specific LLMs or make LLM applications, please contact us at ► [click here to contact](https://www.upstage.ai/private-llm?utm_source=huggingface&utm_medium=link&utm_campaign=privatellm) - As of August 1st, our 70B model has reached the top spot in openLLM rankings, marking itself as the current leading performer globally.