inference: false
language:
- en
license: llama2
model_creator: NousResearch
model_link: https://huggingface.co/NousResearch/Nous-Hermes-Llama2-70b
model_name: Nous Hermes Llama2 70B
model_type: llama
quantized_by: TheBloke
tags:
- llama-2
- self-instruct
- distillation
- synthetic instruction
![TheBlokeAI](https://i.imgur.com/EBdldam.jpg)
TheBloke's LLM work is generously supported by a grant from andreessen horowitz (a16z)
Nous Hermes Llama2 70B - GGML
- Model creator: NousResearch
- Original model: Nous Hermes Llama2 70B
Description
This repo contains GGML format model files for NousResearch's Nous Hermes Llama2 70B.
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, commit
e76d630
and later. - text-generation-webui, the most widely used web UI.
- KoboldCpp, version 1.37 and later. A powerful GGML web UI, especially good for story telling.
- LM Studio, 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, version 0.1.77 and later. A Python library with LangChain support, and OpenAI-compatible API server.
- 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.
- 2, 3, 4, 5, 6 and 8-bit GGUF models for CPU+GPU inference
- 2, 3, 4, 5, 6 and 8-bit GGML models for CPU+GPU inference (deprecated)
- NousResearch's original unquantised fp16 model in pytorch format, for GPU inference and for further conversions
Prompt template: Alpaca-InstructOnly
### Instruction:
{prompt}
### Response:
Compatibility
Requires llama.cpp commit e76d630
or later.
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 |
---|---|---|---|---|---|
nous-hermes-llama2-70b.ggmlv3.Q2_K.bin | Q2_K | 2 | 28.96 GB | 31.46 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. |
nous-hermes-llama2-70b.ggmlv3.Q3_K_S.bin | Q3_K_S | 3 | 30.09 GB | 32.59 GB | New k-quant method. Uses GGML_TYPE_Q3_K for all tensors |
nous-hermes-llama2-70b.ggmlv3.Q3_K_M.bin | Q3_K_M | 3 | 33.39 GB | 35.89 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 |
nous-hermes-llama2-70b.ggmlv3.Q3_K_L.bin | Q3_K_L | 3 | 36.49 GB | 38.99 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 |
nous-hermes-llama2-70b.ggmlv3.Q4_0.bin | Q4_0 | 4 | 38.80 GB | 41.30 GB | Original quant method, 4-bit. |
nous-hermes-llama2-70b.ggmlv3.Q4_K_S.bin | Q4_K_S | 4 | 39.18 GB | 41.68 GB | New k-quant method. Uses GGML_TYPE_Q4_K for all tensors |
nous-hermes-llama2-70b.ggmlv3.Q4_K_M.bin | Q4_K_M | 4 | 41.69 GB | 44.19 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 |
nous-hermes-llama2-70b.ggmlv3.Q4_1.bin | Q4_1 | 4 | 43.12 GB | 45.62 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. |
nous-hermes-llama2-70b.ggmlv3.Q5_0.bin | Q5_0 | 5 | 47.43 GB | 49.93 GB | Original quant method, 5-bit. Higher accuracy, higher resource usage and slower inference. |
nous-hermes-llama2-70b.ggmlv3.Q5_K_S.bin | Q5_K_S | 5 | 47.74 GB | 50.24 GB | New k-quant method. Uses GGML_TYPE_Q5_K for all tensors |
nous-hermes-llama2-70b.ggmlv3.Q5_K_M.bin | Q5_K_M | 5 | 49.03 GB | 51.53 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
I use the following command line; adjust for your tastes and needs:
./main -t 10 -ngl 40 -gqa 8 -m nous-hermes-llama2-70b.ggmlv3.q4_K_M.bin --color -c 4096 --temp 0.7 --repeat_penalty 1.1 -n -1 -p "### Instruction:\n\nWrite a story about llamas\n\n### Response:"
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 <PROMPT>
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
How to run in text-generation-webui
Further instructions here: text-generation-webui/docs/llama.cpp-models.md.
Discord
For further support, and discussions on these models and AI in general, join us at:
Thanks, and how to contribute.
Thanks to the 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: Sam, theTransient, Jonathan Leane, Steven Wood, webtim, Johann-Peter Hartmann, Geoffrey Montalvo, Gabriel Tamborski, Willem Michiel, John Villwock, Derek Yates, Mesiah Bishop, Eugene Pentland, Pieter, Chadd, Stephen Murray, Daniel P. Andersen, terasurfer, Brandon Frisco, Thomas Belote, Sid, Nathan LeClaire, Magnesian, Alps Aficionado, Stanislav Ovsiannikov, Alex, Joseph William Delisle, Nikolai Manek, Michael Davis, Junyu Yang, K, J, Spencer Kim, Stefan Sabev, Olusegun Samson, transmissions 11, Michael Levine, Cory Kujawski, Rainer Wilmers, zynix, Kalila, Luke @flexchar, Ajan Kanaga, Mandus, vamX, Ai Maven, Mano Prime, Matthew Berman, subjectnull, Vitor Caleffi, Clay Pascal, biorpg, alfie_i, 阿明, Jeffrey Morgan, ya boyyy, Raymond Fosdick, knownsqashed, Olakabola, Leonard Tan, ReadyPlayerEmma, Enrico Ros, Dave, Talal Aujan, Illia Dulskyi, Sean Connelly, senxiiz, Artur Olbinski, Elle, Raven Klaugh, Fen Risland, Deep Realms, Imad Khwaja, Fred von Graf, Will Dee, usrbinkat, SuperWojo, Alexandros Triantafyllidis, Swaroop Kallakuri, Dan Guido, John Detwiler, Pedro Madruga, Iucharbius, Viktor Bowallius, Asp the Wyvern, Edmond Seymore, Trenton Dambrowitz, Space Cruiser, Spiking Neurons AB, Pyrater, LangChain4j, Tony Hughes, Kacper Wikieł, Rishabh Srivastava, David Ziegler, Luke Pendergrass, Andrey, Gabriel Puliatti, Lone Striker, Sebastain Graf, Pierre Kircher, Randy H, NimbleBox.ai, Vadim, danny, Deo Leter
Thank you to all my generous patrons and donaters!
And thank you again to a16z for their generous grant.
Original model card: NousResearch's Nous Hermes Llama2 70B
Model Card: Nous-Hermes-Llama2-70b
Compute provided by PygmalionAI, thank you! Follow PygmalionAI on Twitter @pygmalion_ai.
Model Description
Nous-Hermes-Llama2-70b is a state-of-the-art language model fine-tuned on over 300,000 instructions. This model was fine-tuned by Nous Research, with Teknium and Emozilla leading the fine tuning process and dataset curation, Pygmalion sponsoring the compute, and several other contributors.
This Hermes model uses the exact same dataset as Hermes on Llama-1. This is to ensure consistency between the old Hermes and new, for anyone who wanted to keep Hermes as similar to the old one, just more capable.
This model stands out for its long responses, lower hallucination rate, and absence of OpenAI censorship mechanisms in the synthetic training data. The fine-tuning process was performed with a 4096 sequence length on an 8x H100 80GB machine.
Model Training
The model was trained almost entirely on synthetic GPT-4 outputs. Curating high quality GPT-4 datasets enables incredibly high quality in knowledge, task completion, and style.
This includes data from diverse sources such as GPTeacher, the general, roleplay v1&2, code instruct datasets, Nous Instruct & PDACTL (unpublished), and several others, detailed further below
Collaborators
The model fine-tuning and the datasets were a collaboration of efforts and resources between Teknium, Karan4D, Emozilla, Huemin Art, and Pygmalion AI.
Special mention goes to @winglian for assisting in some of the training issues.
Huge shoutout and acknowledgement is deserved for all the dataset creators who generously share their datasets openly.
Among the contributors of datasets:
- GPTeacher was made available by Teknium
- Wizard LM by nlpxucan
- Nous Research Instruct Dataset was provided by Karan4D and HueminArt.
- GPT4-LLM and Unnatural Instructions were provided by Microsoft
- Airoboros dataset by jondurbin
- Camel-AI's domain expert datasets are from Camel-AI
- CodeAlpaca dataset by Sahil 2801.
If anyone was left out, please open a thread in the community tab.
Prompt Format
The model follows the Alpaca prompt format:
### Instruction:
<prompt>
### Response:
<leave a newline blank for model to respond>
or
### Instruction:
<prompt>
### Input:
<additional context>
### Response:
<leave a newline blank for model to respond>
Benchmarks:
GPT4All Suite:
hf-causal-experimental (pretrained=/home/data/axolotl/Nous-Hermes-Llama2-70b,dtype=float16,use_accelerate=True), limit: None, provide_description: False, num_fewshot: 0, batch_size: None
| Task |Version| Metric |Value | |Stderr|
|-------------|------:|--------|-----:|---|-----:|
|arc_challenge| 0|acc |0.5734|± |0.0145|
| | |acc_norm|0.6015|± |0.0143|
|arc_easy | 0|acc |0.8422|± |0.0075|
| | |acc_norm|0.8253|± |0.0078|
|boolq | 1|acc |0.8422|± |0.0064|
|hellaswag | 0|acc |0.6519|± |0.0048|
| | |acc_norm|0.8363|± |0.0037|
|openbookqa | 0|acc |0.3880|± |0.0218|
| | |acc_norm|0.5000|± |0.0224|
|piqa | 0|acc |0.8313|± |0.0087|
| | |acc_norm|0.8351|± |0.0087|
|winogrande | 0|acc |0.7751|± |0.0117|
BigBench Suite:
hf-causal-experimental (pretrained=/home/data/axolotl/Nous-Hermes-Llama2-70b,dtype=float16,use_accelerate=True), limit: None, provide_description: False, num_fewshot: 0, batch_size: None
| Task |Version| Metric |Value | |Stderr|
|------------------------------------------------|------:|---------------------|-----:|---|-----:|
|bigbench_causal_judgement | 0|multiple_choice_grade|0.6579|± |0.0345|
|bigbench_date_understanding | 0|multiple_choice_grade|0.7344|± |0.0230|
|bigbench_disambiguation_qa | 0|multiple_choice_grade|0.3023|± |0.0286|
|bigbench_geometric_shapes | 0|multiple_choice_grade|0.2340|± |0.0224|
| | |exact_str_match |0.0000|± |0.0000|
|bigbench_logical_deduction_five_objects | 0|multiple_choice_grade|0.2760|± |0.0200|
|bigbench_logical_deduction_seven_objects | 0|multiple_choice_grade|0.1871|± |0.0148|
|bigbench_logical_deduction_three_objects | 0|multiple_choice_grade|0.4467|± |0.0288|
|bigbench_movie_recommendation | 0|multiple_choice_grade|0.3240|± |0.0210|
|bigbench_navigate | 0|multiple_choice_grade|0.5000|± |0.0158|
|bigbench_reasoning_about_colored_objects | 0|multiple_choice_grade|0.6605|± |0.0106|
|bigbench_ruin_names | 0|multiple_choice_grade|0.4598|± |0.0236|
|bigbench_salient_translation_error_detection | 0|multiple_choice_grade|0.2585|± |0.0139|
|bigbench_snarks | 0|multiple_choice_grade|0.6630|± |0.0352|
|bigbench_sports_understanding | 0|multiple_choice_grade|0.7394|± |0.0140|
|bigbench_temporal_sequences | 0|multiple_choice_grade|0.4440|± |0.0157|
|bigbench_tracking_shuffled_objects_five_objects | 0|multiple_choice_grade|0.2168|± |0.0117|
|bigbench_tracking_shuffled_objects_seven_objects| 0|multiple_choice_grade|0.1531|± |0.0086|
|bigbench_tracking_shuffled_objects_three_objects| 0|multiple_choice_grade|0.4467|± |0.0288|
AGIEval:
hf-causal-experimental (pretrained=/home/data/axolotl/Nous-Hermes-Llama2-70b,dtype=float16,use_accelerate=True), limit: None, provide_description: False, num_fewshot: 0, batch_size: None
| Task |Version| Metric |Value | |Stderr|
|------------------------------|------:|--------|-----:|---|-----:|
|agieval_aqua_rat | 0|acc |0.2480|± |0.0272|
| | |acc_norm|0.2362|± |0.0267|
|agieval_logiqa_en | 0|acc |0.3917|± |0.0191|
| | |acc_norm|0.3932|± |0.0192|
|agieval_lsat_ar | 0|acc |0.2217|± |0.0275|
| | |acc_norm|0.2000|± |0.0264|
|agieval_lsat_lr | 0|acc |0.5765|± |0.0219|
| | |acc_norm|0.4922|± |0.0222|
|agieval_lsat_rc | 0|acc |0.6914|± |0.0282|
| | |acc_norm|0.6022|± |0.0299|
|agieval_sat_en | 0|acc |0.8641|± |0.0239|
| | |acc_norm|0.8204|± |0.0268|
|agieval_sat_en_without_passage| 0|acc |0.5291|± |0.0349|
| | |acc_norm|0.4709|± |0.0349|
|agieval_sat_math | 0|acc |0.4136|± |0.0333|
| | |acc_norm|0.3455|± |0.0321|
Resources for Applied Use Cases:
Check out LM Studio for a nice chatgpt style interface here: https://lmstudio.ai/
For an example of a back and forth chatbot using huggingface transformers and discord, check out: https://github.com/teknium1/alpaca-discord
For an example of a roleplaying discord chatbot, check out this: https://github.com/teknium1/alpaca-roleplay-discordbot
Future Plans
We plan to continue to iterate on both more high quality data, and new data filtering techniques to eliminate lower quality data going forward.
Model Usage
The model is available for download on Hugging Face. It is suitable for a wide range of language tasks, from generating creative text to understanding and following complex instructions.
Training procedure
The following bitsandbytes
quantization config was used during training:
- quant_method: bitsandbytes
- load_in_8bit: False
- load_in_4bit: True
- llm_int8_threshold: 6.0
- llm_int8_skip_modules: None
- llm_int8_enable_fp32_cpu_offload: False
- llm_int8_has_fp16_weight: False
- bnb_4bit_quant_type: nf4
- bnb_4bit_use_double_quant: True
- bnb_4bit_compute_dtype: bfloat16
The following bitsandbytes
quantization config was used during training:
- quant_method: bitsandbytes
- load_in_8bit: False
- load_in_4bit: True
- llm_int8_threshold: 6.0
- llm_int8_skip_modules: None
- llm_int8_enable_fp32_cpu_offload: False
- llm_int8_has_fp16_weight: False
- bnb_4bit_quant_type: nf4
- bnb_4bit_use_double_quant: True
- bnb_4bit_compute_dtype: bfloat16
Framework versions
PEFT 0.5.0.dev0
PEFT 0.5.0.dev0