VMware's Open Llama 7B v2 Open Instruct GGML
These files are GGML format model files for VMware's Open Llama 7B v2 Open Instruct.
GGML files are for CPU + GPU inference using llama.cpp and libraries and UIs which support this format, such as:
- KoboldCpp, a powerful GGML web UI with full GPU acceleration out of the box. Especially good for story telling.
- LoLLMS Web UI, a great web UI with GPU acceleration via the c_transformers backend.
- LM Studio, a fully featured local GUI. Supports full GPU accel on macOS. Also supports Windows, without GPU accel.
- text-generation-webui, the most popular web UI. Requires extra steps to enable GPU accel via llama.cpp backend.
- ctransformers, a Python library with LangChain support and OpenAI-compatible AI server.
- llama-cpp-python, a Python library with OpenAI-compatible API server.
These files were quantised using hardware kindly provided by Latitude.sh.
Repositories available
- GPTQ models for GPU inference, with multiple quantisation parameter options.
- 2, 3, 4, 5, 6 and 8-bit GGML models for CPU+GPU inference
- Unquantised fp16 model in pytorch format, for GPU inference and for further conversions
Prompt template: Alpaca
Below is an instruction that describes a task. Write a response that appropriately completes the request.
### Instruction: {prompt}
### Response:
Compatibility
Original llama.cpp quant methods: q4_0, q4_1, q5_0, q5_1, q8_0
These are guaranteed to be compatible with any UIs, tools and libraries released since late May. They may be phased out soon, as they are largely superseded by the new k-quant methods.
New k-quant methods: q2_K, q3_K_S, q3_K_M, q3_K_L, q4_K_S, q4_K_M, q5_K_S, q6_K
These new quantisation methods are compatible with llama.cpp as of June 6th, commit 2d43387.
They are now also compatible with recent releases of text-generation-webui, KoboldCpp, llama-cpp-python, ctransformers, rustformers and most others. For compatibility with other tools and libraries, please check their documentation.
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 |
|---|---|---|---|---|---|
| open-llama-7b-v2-open-instruct.ggmlv3.q2_K.bin | q2_K | 2 | 2.87 GB | 5.37 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. |
| open-llama-7b-v2-open-instruct.ggmlv3.q3_K_L.bin | q3_K_L | 3 | 3.60 GB | 6.10 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 |
| open-llama-7b-v2-open-instruct.ggmlv3.q3_K_M.bin | q3_K_M | 3 | 3.28 GB | 5.78 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 |
| open-llama-7b-v2-open-instruct.ggmlv3.q3_K_S.bin | q3_K_S | 3 | 2.95 GB | 5.45 GB | New k-quant method. Uses GGML_TYPE_Q3_K for all tensors |
| open-llama-7b-v2-open-instruct.ggmlv3.q4_0.bin | q4_0 | 4 | 3.79 GB | 6.29 GB | Original quant method, 4-bit. |
| open-llama-7b-v2-open-instruct.ggmlv3.q4_1.bin | q4_1 | 4 | 4.21 GB | 6.71 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. |
| open-llama-7b-v2-open-instruct.ggmlv3.q4_K_M.bin | q4_K_M | 4 | 4.08 GB | 6.58 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 |
| open-llama-7b-v2-open-instruct.ggmlv3.q4_K_S.bin | q4_K_S | 4 | 3.83 GB | 6.33 GB | New k-quant method. Uses GGML_TYPE_Q4_K for all tensors |
| open-llama-7b-v2-open-instruct.ggmlv3.q5_0.bin | q5_0 | 5 | 4.63 GB | 7.13 GB | Original quant method, 5-bit. Higher accuracy, higher resource usage and slower inference. |
| open-llama-7b-v2-open-instruct.ggmlv3.q5_1.bin | q5_1 | 5 | 5.06 GB | 7.56 GB | Original quant method, 5-bit. Even higher accuracy, resource usage and slower inference. |
| open-llama-7b-v2-open-instruct.ggmlv3.q5_K_M.bin | q5_K_M | 5 | 4.78 GB | 7.28 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 |
| open-llama-7b-v2-open-instruct.ggmlv3.q5_K_S.bin | q5_K_S | 5 | 4.65 GB | 7.15 GB | New k-quant method. Uses GGML_TYPE_Q5_K for all tensors |
| open-llama-7b-v2-open-instruct.ggmlv3.q6_K.bin | q6_K | 6 | 5.53 GB | 8.03 GB | New k-quant method. Uses GGML_TYPE_Q8_K - 6-bit quantization - for all tensors |
| open-llama-7b-v2-open-instruct.ggmlv3.q8_0.bin | q8_0 | 8 | 7.16 GB | 9.66 GB | Original quant method, 8-bit. Almost indistinguishable from float16. High resource use and slow. Not recommended for most users. |
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 32 -m open-llama-7b-v2-open-instruct.ggmlv3.q4_0.bin --color -c 2048 --temp 0.7 --repeat_penalty 1.1 -n -1 -p "### Instruction: Write a story about llamas\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.
Change -ngl 32 to the number of layers to offload to GPU. Remove it if you don't have GPU acceleration.
If you want to have a chat-style conversation, replace the -p <PROMPT> argument with -i -ins
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: Luke from CarbonQuill, Aemon Algiz.
Patreon special mentions: Space Cruiser, Nikolai Manek, Sam, Chris McCloskey, Rishabh Srivastava, Kalila, Spiking Neurons AB, Khalefa Al-Ahmad, WelcomeToTheClub, Chadd, Lone Striker, Viktor Bowallius, Edmond Seymore, Ai Maven, Chris Smitley, Dave, Alexandros Triantafyllidis, Luke @flexchar, Elle, ya boyyy, Talal Aujan, Alex , Jonathan Leane, Deep Realms, Randy H, subjectnull, Preetika Verma, Joseph William Delisle, Michael Levine, chris gileta, K, Oscar Rangel, LangChain4j, Trenton Dambrowitz, Eugene Pentland, Johann-Peter Hartmann, Femi Adebogun, Illia Dulskyi, senxiiz, Daniel P. Andersen, Sean Connelly, Artur Olbinski, RoA, Mano Prime, Derek Yates, Raven Klaugh, David Flickinger, Willem Michiel, Pieter, Willian Hasse, vamX, Luke Pendergrass, webtim, Ghost , Rainer Wilmers, Nathan LeClaire, Will Dee, Cory Kujawski, John Detwiler, Fred von Graf, biorpg, Iucharbius , Imad Khwaja, Pierre Kircher, terasurfer , Asp the Wyvern, John Villwock, theTransient, zynix , Gabriel Tamborski, Fen Risland, Gabriel Puliatti, Matthew Berman, Pyrater, SuperWojo, Stephen Murray, Karl Bernard, Ajan Kanaga, Greatston Gnanesh, Junyu Yang.
Thank you to all my generous patrons and donaters!
Original model card: VMware's Open Llama 7B v2 Open Instruct
VMware/open-llama-7B-v2-open-instruct
Instruction-tuned version of the fully trained Open LLama 7B v2 model. The model is open for COMMERCIAL USE.
- This model performs better on code compared to v1 due to the improvements made on the base model by the openlm-research team.
- The instruction model is trained on an improved instruction tuning dataset compared to v1
NOTE : The model was trained using the Alpaca prompt template
NOTE : Fast tokenizer results in incorrect encoding, set the use_fast = False parameter, when instantiating the tokenizer
License
Commercially Viable
Open-instruct-v1
Mosaic/Dolly-HHRLHF + filtered OASST1 - cc by 3.0
Subset of COT SUBMIX (FROM FLAN V2) Zeroshot examples
- ESNLI - MIT
- ECQA - CDLA 1.0 - Sharing
- Strategy - MIT
- CREAK - MIT
- gsmk8 - MIT
- aqua - MIT
- qasc - Apache 2.0
- Language Model, (openlm-research/open_llama_v2_7b) is under apache-2.0
Nomenclature
- Model : Open-llama-v2
- Model Size: 7B parameters
- Dataset: Open-instruct(oasst,dolly, hhrlhf)
Use in Transformers
import os
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
model_name = 'VMware/open-llama-7b-open-instruct'
tokenizer = AutoTokenizer.from_pretrained(model_name, use_fast=False)
model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype=torch.float16, device_map='sequential')
prompt_template = "Below is an instruction that describes a task. Write a response that appropriately completes the request.\n\n### Instruction:\n{instruction}\n\n### Response:"
prompt = """What is attention mechanism of a transformer model?
Write a python code to illustrate how attention works within a transformer model using numpy library. Donot use pytorch or tensorflow."""
inputt = prompt_template.format(instruction= prompt)
input_ids = tokenizer(inputt, return_tensors="pt").input_ids.to("cuda")
output1 = model.generate(input_ids, max_length=512)
input_length = input_ids.shape[1]
output1 = output1[:, input_length:]
output = tokenizer.decode(output1[0])
print(output)
'''
Sure, I can help you with that!
Attention mechanisms in transformer models are typically implemented using the attention mechanism in the self-attention layer. Self-attention allows the model to focus on different parts of the input sequence when processing it. This is achieved by computing a set of attention weights, which are used to weigh the contribution of each input element to the output.
Here's an example code using NumPy to illustrate how attention works in a transformer model:
```python
import numpy as np
def attention_weights(query, key, value, mask):
# Query, key, and value are input tensors. Mask is a tensor of zeros and ones that represents the attention mask.
# It is used to prevent the model from attending to certain positions in the input sequence if they are not relevant.
# The attention weights are the element-wise product of the query, key, and mask tensors.
# The result is a tensor of the same shape as the query tensor.
# Compute the dot product between the query tensor and the key tensor
dot = np.matmul(query, key)
# Compute the element-wise softmax of the dot product tensor
exp_dot = np.exp(dot)
# Multiply the dot product and the softmax of the dot product tensors
weights = dot * exp_dot
# Return the attention weights as a NumPy tensor
return weights
# Define the input sequence
query = np.array([[0.1, 0.2, 0.3], [0.4, 0.5, 0.6]])
key = np.array([[0.1, 0.2], [0.3, 0.4]])
value = np.array([[0.1, 0.2, 0.3], [0.4, 0.5, 0.6]])
mask = np.array([[False, True, True], [False, True, True]])
# Compute the attention weights
weights = attention_weights(query, key, value, mask)
# Print the attention weights
print(weights)
In this example, the attention_weights function takes as input the query tensor, key tensor, value tensor, and mask tensor. It computes the dot product between the query and key tensors using the np.matmul function, and then applies a softmax function using the np.exp function to the element-wise dot product tensor. It then multiplies the dot product and softmax tensors using the np.matmul function, and returns the result as a NumPy tensor.
The query, key, and value tensors represent the input sequence to the transformer model. The mask tensor represents the attention mask, which is used to prevent the model from attending to certain positions in the input sequence if they are not relevant.
The output of the attention_weights function is a NumPy tensor that represents the attention weights for the input sequence. These weights are used by the transformer model to weigh the contribution of each input element to the output.
I hope this helps! '''
## Finetuning details
The finetuning scripts will be available in our [RAIL Github Repository](https://github.com/vmware-labs/research-and-development-artificial-intelligence-lab/tree/main/instruction-tuning)
## Evaluation
<B>TODO</B>