Document BF16 INT8 conditioning cosine parity
Browse files
README.md
CHANGED
|
@@ -69,6 +69,18 @@ Download these from [Comfy-Org/MiniMax-H3](https://huggingface.co/Comfy-Org/Mini
|
|
| 69 |
|
| 70 |
The showcase used one matched 1344x768, 243-frame, 24 FPS, 25-step T2V contract. The three source outputs and merged comparison are technical assets. They do not establish a visual-quality winner. The recovered 8B showcase uses the BF16 base; the INT8 ConvRot variant has separate numerical conditioning-parity evidence.
|
| 71 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 72 |
The requested hard cuts at 3 and 7 seconds did not pass for any encoder. Read `evidence/cut_diagnostics.json` before using the showcase as prompt-adherence evidence.
|
| 73 |
|
| 74 |
The recovered 8B and pruned-24 variants were validated for text-only T2V. Do not use them for image, first-frame, last-frame, or reference inputs.
|
|
|
|
| 69 |
|
| 70 |
The showcase used one matched 1344x768, 243-frame, 24 FPS, 25-step T2V contract. The three source outputs and merged comparison are technical assets. They do not establish a visual-quality winner. The recovered 8B showcase uses the BF16 base; the INT8 ConvRot variant has separate numerical conditioning-parity evidence.
|
| 71 |
|
| 72 |
+
### Recovered 8B BF16 versus INT8 ConvRot conditioning parity
|
| 73 |
+
|
| 74 |
+
For one fixed seven-token prompt, the recovered 8B BF16 and INT8 ConvRot variants produced conditioning tensors with shape `[1, 7, 5120]`. The output tensors were flattened and compared with cosine similarity:
|
| 75 |
+
|
| 76 |
+
| Metric | Result |
|
| 77 |
+
|---|---:|
|
| 78 |
+
| Cosine similarity | `0.9999650716781616` |
|
| 79 |
+
| Mean absolute difference | `0.015660330653190613` |
|
| 80 |
+
| Maximum absolute difference | `0.6452102661132812` |
|
| 81 |
+
|
| 82 |
+
This is a numerical conditioning-parity diagnostic for one prompt. It is not a visual-quality score, a prompt-adherence score, or proof that the INT8 variant produces identical video results. Full details are in `evidence/release_loader_smoke.json`.
|
| 83 |
+
|
| 84 |
The requested hard cuts at 3 and 7 seconds did not pass for any encoder. Read `evidence/cut_diagnostics.json` before using the showcase as prompt-adherence evidence.
|
| 85 |
|
| 86 |
The recovered 8B and pruned-24 variants were validated for text-only T2V. Do not use them for image, first-frame, last-frame, or reference inputs.
|