license: mit
library_name: transformers
pipeline_tag: text-generation
tags:
- code
- deepseek
- gguf
- bf16
- chinese
- english
metrics:
- accuracy
Deepseek-V2-Chat-GGUF
Quantizised from https://huggingface.co/deepseek-ai/DeepSeek-V2-Chat
Using llama.cpp fork: https://github.com/fairydreaming/llama.cpp/tree/deepseek-v2
TODO: Make llamafile for Q2_K and Q4_K_M
Warning: This will not work unless you compile llama.cpp from the repo provided (and set metadata KV overrides)!
How to use:
Downloading the bf16:
- Find the relevant directory
- Download all files
- Run merge.py
- Merged GGUF should appear
Downloading the quantizations:
- Find the relevant directory
- Download all files
- Point to the first split (most programs should load all the splits automatically now)
Running in llama.cpp:
To start in command line interactive mode (text completion):
main -m DeepSeek-V2-Chat.{quant}.gguf -c {context length} --color -i
To use llama.cpp OpenAI compatible server:
server \
-m DeepSeek-V2-Chat.{quant}.gguf \
-c {context_length} \
(--color [recommended: colored output in supported terminals]) \
(-i [note: interactive mode]) \
(--mlock [note: avoid using swap]) \
(--verbose) \
(--log-disable [note: disable logging to file, may be useful for prod]) \
(--metrics [note: prometheus compatible monitoring endpoint]) \
(--api-key [string]) \
(--port [int]) \
(--flash-attn [note: must be fully offloaded to supported GPU])
Making an importance matrix:
imatrix \
-m DeepSeek-V2-Chat.{quant}.gguf \
-f groups_merged.txt \
--verbosity [0, 1, 2] \
-ngl {GPU offloading; must build with CUDA} \
--ofreq {recommended: 1}
Making a quant:
quantize \
DeepSeek-V2-Chat.bf16.gguf \
DeepSeek-V2-Chat.{quant}.gguf \
{quant} \
(--imatrix [file])
Quants:
- bf16 [size: 439gb]
- q8_0 (later, please use q4_k_m for now) [estimated size: 233.27gb]
- q4_k_m [size: 132gb]
- q2_k [size: 80gb]
- iq2_xxs [size: 61.5gb]
- iq3_xs (uploading) [size: 89.6gb]
- iq1_m [size: 27.3gb]
Note: Use iMatrix quants only if you can fully offload to GPU, otherwise speed will be affected a lot.
Planned Quants (using importance matrix):
- q5_k_m
- q5_k_s
- q3_k_m
- q6_k
- iq4_nl
- iq4_xs
- iq2_xs
- iq2_s
- iq2_m
- iq1_s (note: for fun only, this quant is likely useless)
Note: the model files do not have some DeepSeek v2 specific parameters, will look into adding them
Please use commit 039896407afd40e54321d47c5063c46a52da3e01, otherwise use these metadata KV overrides:
deepseek2.attention.q_lora_rank=int:1536
deepseek2.attention.kv_lora_rank=int:512
deepseek2.expert_shared_count=int:2
deepseek2.expert_feed_forward_length=int:1536
deepseek2.experts_weight_scale=int:16
deepseek2.leading_dense_block_count=int:1
rope.scaling.yarn_log_multiplier=float:0.0707
A precompiled AVX2 version is avaliable at llama.cpp-039896407afd40e54321d47c5063c46a52da3e01.zip in the root of this repo.
License:
- DeepSeek license for model weights
- MIT license for any repo code
Performance:
~1.5t/s with Ryzen 3 3700x (96gb 3200mhz) [Q2_K]
iMatrix:
Find imatrix.dat in the root of this repo, made with a Q2_K quant (see here for info: https://github.com/ggerganov/llama.cpp/issues/5153#issuecomment-1913185693)
Using groups_merged.txt, find it here: https://github.com/ggerganov/llama.cpp/discussions/5263#discussioncomment-8395384
Censorship:
This model is quite censored, finetuning on toxic DPO might help.