52 lines
1.6 KiB
Markdown
52 lines
1.6 KiB
Markdown
---
|
|
license: apache-2.0
|
|
library_name: transformers
|
|
pipeline_tag: text-generation
|
|
base_model:
|
|
- mistralai/Mistral-Nemo-Instruct-2407
|
|
language:
|
|
- en
|
|
- ru
|
|
---
|
|
|
|
# Vikra-HCT-YeAM-Vikhr-NemoGemma-12B_plus_1B
|
|
|
|
HCT architecture release. YeAM (Yet Another Merge) implementation invariant.
|
|
|
|
## What it is
|
|
|
|
A large (12B-class) checkpoint produced via HCT-compatible merging with 1B Gemma.
|
|
Published in standard Hugging Face format (safetensors + sharded index) and intended to be convertible to GGUF.
|
|
|
|
## YeAM summary
|
|
|
|
YeAM performs a controlled merge in a real 4D geometric formulation with ray-intersection alignment in parameter space.
|
|
It also supports targeted knowledge injection (distillation-style) into a chosen model while remaining HF-compatible.
|
|
|
|
## Usage (Transformers)
|
|
|
|
```python
|
|
from transformers import AutoTokenizer, AutoModelForCausalLM
|
|
import torch
|
|
|
|
m = "/path/to/Vikra-HCT-YeAM-Vikhr-NemoGemma-12B_plus_1B"
|
|
|
|
tok = AutoTokenizer.from_pretrained(m, use_fast=False)
|
|
model = AutoModelForCausalLM.from_pretrained(
|
|
m,
|
|
torch_dtype=torch.bfloat16,
|
|
device_map="auto",
|
|
).eval()
|
|
|
|
inputs = tok("Привет!", return_tensors="pt").to(model.device)
|
|
out = model.generate(**inputs, max_new_tokens=256)
|
|
print(tok.decode(out[0], skip_special_tokens=True))
|
|
```
|
|
|
|
## GGUF (example)
|
|
|
|
```bash
|
|
python3 /path/to/llama.cpp/convert_hf_to_gguf.py /path/to/model --outtype bf16 --outfile model.bf16.gguf
|
|
/path/to/llama.cpp/build/bin/llama-quantize model.bf16.gguf model.Q6_K.gguf Q6_K
|
|
CUDA_VISIBLE_DEVICES=0,1 /path/to/llama.cpp/build/bin/llama-server -m model.Q6_K.gguf --n-gpu-layers 99 --split-mode layer --tensor-split 1,1
|