|
|
--- |
|
|
library_name: transformers |
|
|
tags: |
|
|
- reasoning |
|
|
- thinking |
|
|
- cognitivecomputations |
|
|
- r1 |
|
|
- llama 3.1 |
|
|
- llama-3 |
|
|
- llama3 |
|
|
- llama-3.1 |
|
|
- cot |
|
|
- deepseek |
|
|
- Llama 3.1 |
|
|
- Hermes |
|
|
- DeepHermes |
|
|
- 1,000,000 context |
|
|
- fine tune |
|
|
- merge |
|
|
- llama-cpp |
|
|
- gguf-my-repo |
|
|
base_model: DavidAU/Llama-3.1-1-million-ctx-DeepHermes-Deep-Reasoning-8B |
|
|
--- |
|
|
|
|
|
# Triangle104/Llama-3.1-1-million-ctx-DeepHermes-Deep-Reasoning-8B-Q8_0-GGUF |
|
|
This model was converted to GGUF format from [`DavidAU/Llama-3.1-1-million-ctx-DeepHermes-Deep-Reasoning-8B`](https://huggingface.co/DavidAU/Llama-3.1-1-million-ctx-DeepHermes-Deep-Reasoning-8B) using llama.cpp via the ggml.ai's [GGUF-my-repo](https://huggingface.co/spaces/ggml-org/gguf-my-repo) space. |
|
|
Refer to the [original model card](https://huggingface.co/DavidAU/Llama-3.1-1-million-ctx-DeepHermes-Deep-Reasoning-8B) for more details on the model. |
|
|
|
|
|
--- |
|
|
Context : 1,000,000 tokens. |
|
|
|
|
|
|
|
|
Required: Llama 3 Instruct template. |
|
|
|
|
|
|
|
|
The Deep Hermes 8B Preview model (reasoning), [ https://huggingface.co/NousResearch/DeepHermes-3-Llama-3-8B-Preview ] |
|
|
converted to 1 million context using Nvidia's Ultra Long 1 million 8B Instruct model. |
|
|
|
|
|
|
|
|
The goal of this model was to stablize long generation and long context "needle in a haystack" issues. |
|
|
|
|
|
|
|
|
According to Nvidia there is both a bump in general performance, as well as perfect "recall" over the entire 1 million context. |
|
|
|
|
|
|
|
|
[ https://huggingface.co/nvidia/Llama-3.1-8B-UltraLong-1M-Instruct ] |
|
|
|
|
|
|
|
|
Additional changes: |
|
|
|
|
|
|
|
|
Model appears to be de-censored / more de-censored. |
|
|
Output generation is improved. |
|
|
Creative output generation is vastly improved. |
|
|
|
|
|
|
|
|
NOTE: Higher temps will result in deeper, richer "thoughts"... and frankly more interesting ones too. |
|
|
|
|
|
|
|
|
The "thinking/reasoning" tech (for the model at this repo) is from the original Llama 3.1 "DeepHermes" model from NousResearch: |
|
|
|
|
|
|
|
|
[ https://huggingface.co/NousResearch/DeepHermes-3-Llama-3-8B-Preview ] |
|
|
|
|
|
--- |
|
|
## Use with llama.cpp |
|
|
Install llama.cpp through brew (works on Mac and Linux) |
|
|
|
|
|
```bash |
|
|
brew install llama.cpp |
|
|
|
|
|
``` |
|
|
Invoke the llama.cpp server or the CLI. |
|
|
|
|
|
### CLI: |
|
|
```bash |
|
|
llama-cli --hf-repo Triangle104/Llama-3.1-1-million-ctx-DeepHermes-Deep-Reasoning-8B-Q8_0-GGUF --hf-file llama-3.1-1-million-ctx-deephermes-deep-reasoning-8b-q8_0.gguf -p "The meaning to life and the universe is" |
|
|
``` |
|
|
|
|
|
### Server: |
|
|
```bash |
|
|
llama-server --hf-repo Triangle104/Llama-3.1-1-million-ctx-DeepHermes-Deep-Reasoning-8B-Q8_0-GGUF --hf-file llama-3.1-1-million-ctx-deephermes-deep-reasoning-8b-q8_0.gguf -c 2048 |
|
|
``` |
|
|
|
|
|
Note: You can also use this checkpoint directly through the [usage steps](https://github.com/ggerganov/llama.cpp?tab=readme-ov-file#usage) listed in the Llama.cpp repo as well. |
|
|
|
|
|
Step 1: Clone llama.cpp from GitHub. |
|
|
``` |
|
|
git clone https://github.com/ggerganov/llama.cpp |
|
|
``` |
|
|
|
|
|
Step 2: Move into the llama.cpp folder and build it with `LLAMA_CURL=1` flag along with other hardware-specific flags (for ex: LLAMA_CUDA=1 for Nvidia GPUs on Linux). |
|
|
``` |
|
|
cd llama.cpp && LLAMA_CURL=1 make |
|
|
``` |
|
|
|
|
|
Step 3: Run inference through the main binary. |
|
|
``` |
|
|
./llama-cli --hf-repo Triangle104/Llama-3.1-1-million-ctx-DeepHermes-Deep-Reasoning-8B-Q8_0-GGUF --hf-file llama-3.1-1-million-ctx-deephermes-deep-reasoning-8b-q8_0.gguf -p "The meaning to life and the universe is" |
|
|
``` |
|
|
or |
|
|
``` |
|
|
./llama-server --hf-repo Triangle104/Llama-3.1-1-million-ctx-DeepHermes-Deep-Reasoning-8B-Q8_0-GGUF --hf-file llama-3.1-1-million-ctx-deephermes-deep-reasoning-8b-q8_0.gguf -c 2048 |
|
|
``` |
|
|
|
