Views
No views yet
1name: Q2.5-ColdBrew-R1-Indigo
2const_tag: &scale_factor 0.7071067812 # 1/sqrt(2) scaling for stability
3
4attenuate-env: &attenuated_env
5 parameters:
6 scale:
7 - filter: q_proj
8 value: *scale_factor
9 - filter: k_proj
10 value: *scale_factor
11 - value: 1.0
12
13slices:
14 - sources:
15 - model: Theros/Qwen2.5-ColdBrew-R1
16 layer_range: [0, 8] # Retaining foundational knowledge and language structure.
17
18 - sources:
19 - model: Theros/Qwen2.5-ColdBrew-R1
20 layer_range: [9, 19] # Full-strength duplication of mid-range reasoning layers.
21
22 - sources:
23 - model: Theros/Qwen2.5-ColdBrew-R1
24 layer_range: [10, 19] # Targeted reinforcement, slightly attenuated to avoid over-dominance.
25 <<: *attenuated_env
26
27 - sources:
28 - model: Theros/Qwen2.5-ColdBrew-R1
29 layer_range: [20, 28] # Keeping higher-level abstract processing untouched for stability.
30
31merge_method: passthrough
32dtype: bfloat16
33normalize: true
34int8_mask: true1!pip install -qU transformers accelerate
2
3from transformers import AutoTokenizer
4import transformers
5import torch
6
7model = "SvalTek/Q2.5-ColdBrew-R1-Indigo"
8messages = [{"role": "user", "content": "What is a large language model?"}]
9
10tokenizer = AutoTokenizer.from_pretrained(model)
11prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
12pipeline = transformers.pipeline(
13 "text-generation",
14 model=model,
15 torch_dtype=torch.float16,
16 device_map="auto",
17)
18
19outputs = pipeline(prompt, max_new_tokens=256, do_sample=True, temperature=0.7, top_k=50, top_p=0.95)
20print(outputs[0]["generated_text"])