Upload manifest_pretokenized_holdout.json with huggingface_hub

manifest_pretokenized_holdout.json

@@ -0,0 +1,22 @@
+{
+  "source_parquet": "data/openthoughts-agent-v1/ota_sft_holdout.parquet",
+  "tokenizer": "data/Qwen3.5-27B-tokenizer",
+  "rows_in": 200,
+  "rows_out": 200,
+  "dropped": {
+    "contract": 0,
+    "too_long": 0,
+    "no_trained_tokens": 0,
+    "error": 0
+  },
+  "total_tokens": 1617384,
+  "trained_tokens": 486811,
+  "trained_fraction": 0.301,
+  "max_seq_len": 32768,
+  "mask_source": "slime/utils/mask_utils.py::gen_multi_turn_loss_mask_qwen3_5 (ported)",
+  "schema": {
+    "input_ids": "list[int] \u2014 the exact tokens, whole-conversation render",
+    "loss_mask": "list[int] \u2014 1 = train on this token, 0 = context only"
+  },
+  "note": "loss_mask is aligned 1:1 with input_ids: mask[i] refers to token i, with no offset applied. To get `labels`, set labels[i] = input_ids[i] where loss_mask[i]==1 else -100, and do NOT shift \u2014 every HuggingFace CausalLM (and Liger's fused CE) shifts internally, so shifting here misaligns the supervision by one token. Shift only if you hand-write the cross-entropy against unshifted logits, in which case shift logits and labels together as usual."
+}