forked from sambanova/ai-starter-kit
-
Notifications
You must be signed in to change notification settings - Fork 0
/
config.yaml
30 lines (27 loc) · 2.29 KB
/
config.yaml
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
llm:
api: "sncloud" # set either sncloud or sambastudio
max_tokens_to_generate: 500
temperature: 0.0
coe: True
select_expert: "llama3-405b"
use_cases:
- General Assistant
- Document Search
- Product Selection
- Code Generation
- Summarization
- Question & Answering
- Query decomposition
models:
Llama2:
Model Architecture: "Advanced architecture with Grouped-Query Attention, optimized for alignment with human preferences."
Architecture Prompting Implications: "Prompts should be structured to take advantage of its supervised fine-tuning and reinforcement learning from human feedback. Prompts that are more aligned with human conversational styles and preferences work well."
Heuristic Guidance: "Llama 2 responds effectively to concise system prompts that direct its persona or response boundaries. This includes prompts like \"Act as if…\", \"You are…\", \"Always/Never…\", or \"Speak like…\". "
References: "https://replicate.com/blog/how-to-prompt-llama\nhttps://arxiv.org/pdf/2307.09288.pdf\nhttps://huggingface.co/blog/llama2#how-to-prompt-llama-2\nhttps://huggingface.co/docs/transformers/main/model_doc/llama2"
Meta Tag Format: "[INST] <<SYS>> \n{system_prompt} \n<</SYS>> \n{user_prompt} \n[/INST]"
Llama3:
Model Architecture: "Advanced architecture with Grouped-Query Attention, optimized for alignment with human preferences."
Architecture Prompting Implications: "Prompts should be structured to take advantage of its supervised fine-tuning and reinforcement learning from human feedback. Prompts that are more aligned with human conversational styles and preferences work well."
Heuristic Guidance: "Llama 3 responds effectively to concise system prompts that direct its persona or response boundaries. This includes prompts like \"Act as if…\", \"You are…\", \"Always/Never…\", or \"Speak like…\". "
References: "https://llama.meta.com/docs/model-cards-and-prompt-formats/meta-llama-3/\nhttps://ai.meta.com/blog/meta-llama-3/\nhttps://arxiv.org/pdf/2407.21783\nhttps://huggingface.co/meta-llama/Meta-Llama-3-8B"
Meta Tag Format: "<|begin_of_text|><|start_header_id|>system<|end_header_id|>{system_prompt}<|eot_id|> \n<|start_header_id|>user<|end_header_id|>{user_prompt}<|eot_id|> \n<|start_header_id|>assistant<|end_header_id|>"