Views
No views yet
flowcast-v4-cascade-qwenSmall hot path. Full automation when you need it.
| Gate | v3 | B2 cascade | Δ |
|---|---|---|---|
| Core hard quality (117) | 100% | 100% | — |
| Expanded hard quality (170) | 99.4% | 99.4% | — |
| Core p50 latency | ~1002ms | ~999ms | ~same |
| Hot path disk | 2.5 GB | ~2.26 GB | Qwen only |
pip install mlx-lm huggingface_hub1from huggingface_hub import snapshot_download
2from gemmaflow_tune.production import create_cascade_runner
3
4light = snapshot_download("nsalerni/gemma-4-e2b-flowcast-v4-cascade-qwen")
5heavy = snapshot_download("nsalerni/gemma-4-e2b-flowcast-v3")
6
7runner = create_cascade_runner(
8 light_adapter_path=light,
9 heavy_adapter_path=heavy,
10)
11runner.load()
12
13# Intent/dictation routes to Qwen; automation lazy-loads v3 E2B
14result = runner.generate("", user_content="...", suite="intent_classification", max_tokens=64)
15print(result.text)| File | Description |
|---|---|
adapters.safetensors | Qwen3 light LoRA weights |
adapter_config.json | LoRA config + heavy adapter reference |
inference_config.json | Cascade runtime settings + benchmark scores |
1{
2 "runner_kind": "cascade",
3 "prompt_mode": "hybrid_slim",
4 "json_early_stop": true,
5 "use_prompt_kv_cache": true,
6 "temperature": 0.02,
7 "top_p": 0.85
8}