Add FP8 to FP16 conversion for MPS compatibility

[kkphyo999-ops]

Summary

• Convert FP8 tensors to FP16 during model loading (both safetensors and .pt/.ckpt paths) when running on Apple MPS devices, preventing runtime errors
• Add fallback FP8 safety checks in cast_to to catch any FP8 tensors that slip through at inference time
• MPS does not support FP8 dtypes natively; conversion goes FP8 → FP32 → FP16

Test plan

• Load an FP8 safetensors model on an MPS (Apple Silicon) device and verify it converts to FP16 with log output
• Load an FP8 .pt/.ckpt model on MPS and verify batch conversion with summary log
• Verify non-MPS (CUDA/CPU) loading is unaffected
• Verify non-FP8 models on MPS load without unnecessary conversions

🤖 Generated with Claude Code

API Node PR Checklist

Scope

• Is API Node Change

Pricing & Billing

• Need pricing update
• No pricing update

If Need pricing update:

• Metronome rate cards updated
• Auto‑billing tests updated and passing

QA

• QA done
• QA not required

Comms

• Informed Kosinkadink

[coderabbitai]

Caution

Review failed

The pull request is closed.

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: f4f6b113-2866-4d89-9f70-22a39d16f7fb

📥 Commits

Reviewing files that changed from the base of the PR and between f6c162d and e419bfa.

⛔ Files ignored due to path filters (1)

• comfy_api_nodes/apis/kling.py is excluded by !comfy_api_nodes/apis/**

📒 Files selected for processing (5)

• comfy/model_management.py
• comfy/utils.py
• comfy_api_nodes/nodes_kling.py
• comfyui_version.py
• pyproject.toml

---

📝 Walkthrough

Walkthrough

This PR makes three independent sets of changes. First, FP8 tensor coercion for MPS devices is added: cast_to in model_management.py now detects FP8 input dtype or weight dtype targeting MPS and downcasts to torch.float16 with a warning; load_torch_file in utils.py applies the same conversion in both the safetensors and torch.load branches. Second, KlingVideoNode gains kling-3.0-turbo support via new helper functions (build_turbo_shot_prompt, execute_kling_turbo), updated model/resolution/pricing schema, and an early-exit execution branch. Third, the project version is bumped from 0.25.0 to 0.25.1 in comfyui_version.py and pyproject.toml.

✨ Finishing Touches

⚔️ Resolve merge conflicts

• Resolve merge conflict in branch mps-fp8-fix

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

Warning

⚠️ This pull request shows signs of AI-generated slop (description_diff_mismatch). It has been flagged by CodeRabbit slop detection and should be reviewed carefully.