models.yaml

# Model preset registry for dev and Hugging Face Space.
# Select active preset with ACTIVE_MODEL; override any field via .env (see .env.example).

defaults:
  # active_model: minicpm-v-4.6
  active_model: minicpm5-1b

  # Default true for local dev (dropdown in Gradio). Space: set ALLOW_MODEL_SWITCH=false.
  allow_model_switch: true

models:
  minicpm-v-4.6:
    label: MiniCPM-V 4.6 (Transformers, ~0.8B, default)
    backend: transformers
    model_id: openbmb/MiniCPM-V-4.6
    trust_remote_code: true
    multimodal: true

  minicpm-v-4.6-gguf:
    label: MiniCPM-V 4.6 (GGUF / llama.cpp)
    backend: llama_cpp
    model_repo: openbmb/MiniCPM-V-4.6-gguf
    model_file: MiniCPM-V-4_6-Q4_K_M.gguf
    multimodal: true
    n_ctx: 8192
    n_gpu_layers: 0

  qwen3b-gguf:
    label: Qwen 2.5 3B Instruct (GGUF)
    backend: llama_cpp
    model_repo: Qwen/Qwen2.5-3B-Instruct-GGUF
    model_file: qwen2.5-3b-instruct-q4_k_m.gguf
    n_ctx: 4096
    n_gpu_layers: 0

  llama32-3b-gguf:
    label: Llama 3.2 3B Instruct (GGUF)
    backend: llama_cpp
    model_repo: bartowski/Llama-3.2-3B-Instruct-GGUF
    model_file: Llama-3.2-3B-Instruct-Q4_K_M.gguf
    n_ctx: 4096
    n_gpu_layers: 0

  minicpm5-1b:
    label: MiniCPM5 1B (Transformers)
    backend: transformers
    model_id: openbmb/MiniCPM5-1B
    trust_remote_code: true

  gemma4-e2b-mobile:
    label: Gemma 4 E2B IT QAT Mobile (Transformers)
    backend: transformers
    model_id: google/gemma-4-E2B-it-qat-mobile-transformers
    trust_remote_code: true

  gemma-merged-local:
    label: Fine-tuned merged model (local path)
    backend: transformers
    model_id: ./gemma_merged_model

  gemma-lora-local:
    label: Fine-tuned LoRA adapter (local path)
    backend: transformers
    model_id: ./gemma_finetuned_model

  minicpm5-1b-lesson-lora:
    label: MiniCPM5 1B lesson LoRA (local)
    backend: transformers
    model_id: openbmb/MiniCPM5-1B
    adapter_path: ./models/finetuned/minicpm5-1b-lora
    trust_remote_code: true

  minicpm5-1b-lesson-merged:
    label: MiniCPM5 1B lesson merged (local)
    backend: transformers
    model_id: ./models/finetuned/minicpm5-1b-lora-merged
    trust_remote_code: true

  # Well-Tuned track: base MiniCPM5-1B + a Modal-published LoRA adapter pulled
  # straight from the Hub (no local files needed). These point at the repos the
  # finetune pipeline publishes once a job clears its lm-eval gate
  # (research/modal/experiments.yaml -> publish.hub_repo).
  minicpm5-1b-teaching-hub:
    label: MiniCPM5 1B teaching LoRA (Hub, fine-tuned)
    backend: transformers
    model_id: openbmb/MiniCPM5-1B
    adapter_path: MSGEncrypted/minicpm5-1b-teaching-lora
    trust_remote_code: true

  minicpm5-1b-math-hub:
    label: MiniCPM5 1B math LoRA (Hub, fine-tuned)
    backend: transformers
    model_id: openbmb/MiniCPM5-1B
    adapter_path: MSGEncrypted/minicpm5-1b-math-lora
    trust_remote_code: true

  minicpm5-1b-language-lesson-hub:
    label: MiniCPM5 1B language lesson LoRA (FR/AR, Hub)
    backend: transformers
    model_id: openbmb/MiniCPM5-1B
    adapter_path: MSGEncrypted/minicpm5-1b-language-lesson-lora
    trust_remote_code: true

  minicpm5-1b-language-lesson-lora:
    label: MiniCPM5 1B language lesson LoRA (FR/AR, local)
    backend: transformers
    model_id: openbmb/MiniCPM5-1B
    adapter_path: ./models/finetuned/language-lesson-lora
    trust_remote_code: true

  tiny-aya-global:
    label: Tiny Aya Global 3.3B (multilingual coach)
    backend: transformers
    model_id: CohereLabs/tiny-aya-global
    trust_remote_code: true

  tiny-aya-water:
    label: Tiny Aya Water 3.3B (European / Asia-Pacific)
    backend: transformers
    model_id: CohereLabs/tiny-aya-water
    trust_remote_code: true

  tiny-aya-fire:
    label: Tiny Aya Fire 3.3B (South Asian)
    backend: transformers
    model_id: CohereLabs/tiny-aya-fire
    trust_remote_code: true

  tiny-aya-earth:
    label: Tiny Aya Earth 3.3B (West Asian / African)
    backend: transformers
    model_id: CohereLabs/tiny-aya-earth
    trust_remote_code: true