modalities/tutorials/instruction_tuning/configs/text_generation_config.yaml at main · Modalities/modalities · GitHub

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
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
settings:
  referencing_keys:
    sample_key: input_ids
    prediction_key: logits
  model_path: checkpoints/2025-07-02__17-25-05_91a54afb/eid_2025-07-02__17-25-05_91a54afb-seen_steps_438-seen_tokens_43057152-target_steps_584-target_tokens_57409536/converted_checkpoint.pth
  device: 0
  sequence_length: 8192

text_inference_component:
  component_key: inference_component
  variant_key: text
  config:
    device: ${settings.device}
    model:
      instance_key: model_raw
      pass_type: BY_REFERENCE
    tokenizer:
      instance_key: tokenizer
      pass_type: BY_REFERENCE
    sequence_length: ${settings.sequence_length}
    eod_token: <|endoftext|>
    prompt_template: "You are Mody, a helpful assistant trained by the modalities team. Answer friendly and informatively to the user's messages.\nUser: {prompt_input}\nAssistant: "
    temperature: 0

model_raw:
  component_key: model
  variant_key: huggingface_pretrained_model
  config:
    model_type: AutoModelForCausalLM
    model_name: Qwen/Qwen2.5-0.5B
    prediction_key: logits
    huggingface_prediction_subscription_key: logits
    sample_key: input_ids

tokenizer:
  component_key: tokenizer
  variant_key: pretrained_hf_tokenizer
  config:
    pretrained_model_name_or_path: Qwen/Qwen2.5-0.5B
    padding: false
    truncation: false
    special_tokens:
      pad_token: <|endoftext|>
      additional_special_tokens:
        - "<|im_start|>"
        - "<|im_end|>"
        - "<|endoftext|>"