mirror of
https://github.com/storytold/cloud-worker.git
synced 2026-10-09 00:09:43 +00:00
docs: pad markdown table columns for raw-text readability
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This commit is contained in:
Hanashi
+8
-8
@@ -136,14 +136,14 @@ already do. For scale-to-zero economics:
|
||||
|
||||
## Published performance reference points
|
||||
|
||||
| Setup | Config | Time |
|
||||
|---|---|---|
|
||||
| 4x H100 80GB (SGLang TP2+Ulysses2) | 1344×768, 124 fr, 50 steps | 13.3 s |
|
||||
| 8x B300 (SGLang, bf16 / online FP8) | same | 19.0 / 18.0 s |
|
||||
| 4x H200 (SGLang Ulysses4) | same | 75.1 s (53.7 s Cache-DiT) |
|
||||
| 2x RTX 5090 (layerwise offload, ~380 GB host RAM) | same | 560 s |
|
||||
| RTX 4090 Laptop 16 GB (SageAttention) | 960×540, 5 s, 20 steps | 182 s |
|
||||
| RTX 3060 12 GB (heavy offload) | 864×480, 124 fr, 20 steps | <9 min |
|
||||
| Setup | Config | Time |
|
||||
|---------------------------------------------------|----------------------------|---------------------------|
|
||||
| 4x H100 80GB (SGLang TP2+Ulysses2) | 1344×768, 124 fr, 50 steps | 13.3 s |
|
||||
| 8x B300 (SGLang, bf16 / online FP8) | same | 19.0 / 18.0 s |
|
||||
| 4x H200 (SGLang Ulysses4) | same | 75.1 s (53.7 s Cache-DiT) |
|
||||
| 2x RTX 5090 (layerwise offload, ~380 GB host RAM) | same | 560 s |
|
||||
| RTX 4090 Laptop 16 GB (SageAttention) | 960×540, 5 s, 20 steps | 182 s |
|
||||
| RTX 3060 12 GB (heavy offload) | 864×480, 124 fr, 20 steps | <9 min |
|
||||
|
||||
Our single-B200 ComfyUI numbers: see `docs/BENCHMARKS.md`.
|
||||
|
||||
|
||||
+23
-23
@@ -18,29 +18,29 @@ ComfyUI finds both storage locations below via `/ComfyUI/extra_model_paths.yaml`
|
||||
|
||||
## Storage locations on the pod
|
||||
|
||||
| Path | Medium | Notes |
|
||||
|---|---|---|
|
||||
| `/workspace/minimax-h3/models/` | RunPod network volume | **Persistent.** The volume has a ~100 GB quota and is effectively full. |
|
||||
| `/dev/shm/minimax-h3-models/` | tmpfs (RAM, 176 GB) | **EPHEMERAL** — wiped on pod restart. Refill with `bash /workspace/minimax-h3/download-weights-tmpfs.sh` (~10 min). |
|
||||
| Path | Medium | Notes |
|
||||
|---------------------------------|-----------------------|---------------------------------------------------------------------------------------------------------------------|
|
||||
| `/workspace/minimax-h3/models/` | RunPod network volume | **Persistent.** The volume has a ~100 GB quota and is effectively full. |
|
||||
| `/dev/shm/minimax-h3-models/` | tmpfs (RAM, 176 GB) | **EPHEMERAL** — wiped on pod restart. Refill with `bash /workspace/minimax-h3/download-weights-tmpfs.sh` (~10 min). |
|
||||
|
||||
## Downloaded — persistent, `/workspace/minimax-h3/models/`
|
||||
|
||||
| File | Family | Size |
|
||||
|---|---|---|
|
||||
| `diffusion_models/minimax_h3_ref2va_pruned_int8_convrot.safetensors` | **int8 pruned** (ref2va) | 21 GB |
|
||||
| `diffusion_models/minimax_h3_fl2va_pruned_int8_convrot.safetensors` | **int8 pruned** (fl2va) | 21 GB |
|
||||
| `diffusion_models/minimax_h3_ref2va_int8_convrot.safetensors` | **int8** (ref2va) | 34 GB |
|
||||
| `text_encoders/qwen3vl_32b_minimax_h3_nvfp4_awq.safetensors` | text encoder (Qwen3-VL-32B, nvfp4 AWQ) | 16 GB |
|
||||
| `vae/minimax_h3_video_vae_fp16.safetensors` | video VAE fp16 | 5.2 GB |
|
||||
| `vae/minimax_h3_audio_vae_fp32.safetensors` | audio VAE fp32 | 0.6 GB |
|
||||
| File | Family | Size |
|
||||
|----------------------------------------------------------------------|----------------------------------------|--------|
|
||||
| `diffusion_models/minimax_h3_ref2va_pruned_int8_convrot.safetensors` | **int8 pruned** (ref2va) | 21 GB |
|
||||
| `diffusion_models/minimax_h3_fl2va_pruned_int8_convrot.safetensors` | **int8 pruned** (fl2va) | 21 GB |
|
||||
| `diffusion_models/minimax_h3_ref2va_int8_convrot.safetensors` | **int8** (ref2va) | 34 GB |
|
||||
| `text_encoders/qwen3vl_32b_minimax_h3_nvfp4_awq.safetensors` | text encoder (Qwen3-VL-32B, nvfp4 AWQ) | 16 GB |
|
||||
| `vae/minimax_h3_video_vae_fp16.safetensors` | video VAE fp16 | 5.2 GB |
|
||||
| `vae/minimax_h3_audio_vae_fp32.safetensors` | audio VAE fp32 | 0.6 GB |
|
||||
|
||||
## Downloaded — ephemeral tmpfs, `/dev/shm/minimax-h3-models/`
|
||||
|
||||
| File | Family | Size |
|
||||
|---|---|---|
|
||||
| `diffusion_models/minimax_h3_fl2va_int8_convrot.safetensors` | **int8** (fl2va) | 34 GB |
|
||||
| `diffusion_models/minimax_h3_ref2va_bf16.safetensors` | **bf16** (ref2va) | 66 GB |
|
||||
| `diffusion_models/minimax_h3_fl2va_bf16.safetensors` | **bf16** (fl2va) | 66 GB |
|
||||
| File | Family | Size |
|
||||
|--------------------------------------------------------------|-------------------|-------|
|
||||
| `diffusion_models/minimax_h3_fl2va_int8_convrot.safetensors` | **int8** (fl2va) | 34 GB |
|
||||
| `diffusion_models/minimax_h3_ref2va_bf16.safetensors` | **bf16** (ref2va) | 66 GB |
|
||||
| `diffusion_models/minimax_h3_fl2va_bf16.safetensors` | **bf16** (fl2va) | 66 GB |
|
||||
|
||||
The int8 family is split across the two locations only because of the volume
|
||||
quota: ref2va_int8 landed on the volume before it filled; fl2va_int8 and both
|
||||
@@ -51,12 +51,12 @@ byte-verified against the HF API).
|
||||
|
||||
## Not downloaded (exist upstream in the same repo)
|
||||
|
||||
| File | Size | Why skipped |
|
||||
|---|---|---|
|
||||
| `diffusion_models/minimax_h3_{fl2va,ref2va}_pruned_bf16.safetensors` | 40 GB each | a 4th family (pruned, unquantized) — not in the 3 requested families |
|
||||
| `diffusion_models/minimax_h3_{fl2va,ref2va}_pruned_fp8_scaled.safetensors` | 21 GB each | a 5th family — same size class as pruned int8 |
|
||||
| `text_encoders/qwen3vl_32b_minimax_h3_bf16.safetensors` | 52 GB | every official template uses the nvfp4_awq TE; bf16 TE would also not fit the volume quota |
|
||||
| `text_encoders/qwen3vl_32b_minimax_h3_int8_convrot.safetensors` | 27 GB | same |
|
||||
| File | Size | Why skipped |
|
||||
|----------------------------------------------------------------------------|------------|--------------------------------------------------------------------------------------------|
|
||||
| `diffusion_models/minimax_h3_{fl2va,ref2va}_pruned_bf16.safetensors` | 40 GB each | a 4th family (pruned, unquantized) — not in the 3 requested families |
|
||||
| `diffusion_models/minimax_h3_{fl2va,ref2va}_pruned_fp8_scaled.safetensors` | 21 GB each | a 5th family — same size class as pruned int8 |
|
||||
| `text_encoders/qwen3vl_32b_minimax_h3_bf16.safetensors` | 52 GB | every official template uses the nvfp4_awq TE; bf16 TE would also not fit the volume quota |
|
||||
| `text_encoders/qwen3vl_32b_minimax_h3_int8_convrot.safetensors` | 27 GB | same |
|
||||
|
||||
## History / gotchas
|
||||
|
||||
|
||||
Reference in New Issue
Block a user