VideoJev-2B / validation.json
ZhuYuhan's picture
Align backbone tensor dtypes with Qwen and retain FP32 decision head
0213e4c verified
Raw History Blame Contribute Delete
426 Bytes
{
"passed": true,
"scope": "dtype conversion and CPU BF16 loading",
"source_revision": "9d1e46dec73b9103cab3a180067d8c7493df4b1e",
"all_source_hashes_verified": true,
"all_keys_and_shapes_verified": true,
"all_tensor_dtypes_verified": true,
"expected_cast_values_verified": true,
"bf16_load_preserves_designated_fp32": true,
"head_forward_float32_and_finite": true,
"benchmark_results_recomputed": false
}