README.md

---
license: cc-by-nc-4.0
language:
- en
---
# ⚡ExLlamaV2 quant of : [L3-8B-Stheno-v3.3-32K](https://huggingface.co/Sao10K/L3-8B-Stheno-v3.3-32K) 
> [!note]
> ➡️ **Exl2 version :** [0.1.5](https://github.com/turboderp/exllamav2/releases/tag/v0.1.5)<br/>
> ➡️ **Cal. dataset :** Default.<br/>
> 📄 <a href="https://huggingface.co/Meggido/L3-8B-Stheno-v3.3-32K-6.5bpw-h8-exl2/resolve/main/measurement.json" download>Measurement.json</a> file.

Trained with compute from [Backyard.ai](https://backyard.ai/) | Thanks to them and @dynafire for helping me out.

---

Training Details:
<br>Trained at 8K Context -> Expanded to 32K Context with PoSE training.

Dataset Modifications:
<br>\- Further Cleaned up Roleplaying Samples -> Quality Check
<br>\- Removed Low Quality Samples from Manual Check -> Increased Baseline Quality Floor
<br>\- More Creative Writing Samples -> 2x Samples
<br>\- Remade and Refined Detailed Instruct Data

Notes:
<br>\- Training run is much less aggressive than previous Stheno versions.
<br>\- This model works when tested in bf16 with the same configs as within the file.
<br>\- I do not know the effects quantisation has on it.
<br>\- Roleplays pretty well. Feels nice in my opinion.
<br>\- It has some issues on long context understanding and reasoning. Much better vs rope scaling normally though, so that is a plus.
<br>\- Reminder, this isn't a native 32K model. It has it's issues, but it's coherent and working well.

Sanity Check // Needle in a Haystack Results:
<br>\- This is not as complex as RULER or NIAN, but it's a basic evaluator. Some improper train examples had Haystack scores ranging from Red to Orange for most of the extended contexts.
![Results](https://huggingface.co/Sao10K/L3-8B-Stheno-v3.3-32K/resolve/main/haystack.png)

Wandb Run:
![Wandb](https://huggingface.co/Sao10K/L3-8B-Stheno-v3.3-32K/resolve/main/wandb.png)

---

Relevant Axolotl Configurations:
<br>-> Taken from [winglian/Llama-3-8b-64k-PoSE](https://huggingface.co/winglian/Llama-3-8b-64k-PoSE)
<br>\- I tried to find my own configs, hours of tinkering but the one he used worked best, so I stuck to it.
<br>\- 2M Rope Theta had the best loss results during training compared to other values.
<br>\- Leaving it at 500K rope wasn't that much worse, but 4M and 8M Theta made the grad_norm values worsen even if loss drops fast.
<br>\- Mixing in Pretraining Data was a PITA. Made it a lot worse with formatting.
<br>\- Pretraining / Noise made it worse at Haystack too? It wasn't all Green, Mainly Oranges.
<br>\- Improper / Bad Rope Theta shows in Grad_Norm exploding to thousands. It'll drop to low values alright, but it's a scary fast drop even with gradient clipping.

```
sequence_len: 8192
use_pose: true
pose_max_context_len: 32768

overrides_of_model_config:
  rope_theta: 2000000.0
  max_position_embeddings: 32768

  # peft_use_dora: true
adapter: lora
peft_use_rslora: true
lora_model_dir:
lora_r: 256
lora_alpha: 256
lora_dropout: 0.1
lora_target_linear: true
lora_target_modules:
  - gate_proj
  - down_proj
  - up_proj
  - q_proj
  - v_proj
  - k_proj
  - o_proj

warmup_steps: 80
gradient_accumulation_steps: 6
micro_batch_size: 1
num_epochs: 2
optimizer: adamw_bnb_8bit
lr_scheduler: cosine_with_min_lr
learning_rate: 0.00004
lr_scheduler_kwargs:
    min_lr: 0.000004
```