Update model card (new model id) and remove provenance/notice files to match v2 export

LEMONDECIDER_BUILD.json

@@ -1,25 +0,0 @@
-{
-  "schema": "lemondecider-build-receipt/1",
-  "model_name": "lemondecider-4b-merge50-20261015-v2",
-  "display_name": "LemonDecider 4B (Qwen3.5-4B 公开数据双路线 SFT 50/50 合并)",
-  "compiled_utc": "2026-10-15T07:10:31Z",
-  "base": "Qwen/Qwen3.5-4B",
-  "training": {
-    "routes": 2,
-    "data": "public data sources",
-    "description": "two independent SFT routes from a common Qwen3.5-4B initialization, trained on public data sources"
-  },
-  "merge": {
-    "method": "exact 0.5*A + 0.5*B elementwise tensor merge, FP32 mean, one BF16 cast; FP32 readout preserved",
-    "tensors": 724,
-    "post_merge_training": false
-  },
-  "runtime": "lemonjev-qwen-decision-runtime-v2: noncausal_full_attention, pooling=last, T=1",
-  "weights": {
-    "shards": 2,
-    "note": "serialization files byte-identical to the parent merged package of the same weights (sha256 verified)"
-  },
-  "eval_policy": "No benchmark score is claimed by this package.",
-  "v2_of": "lemondecider-4b-merge50-20261015-v1",
-  "v2_change": "machine-identifier spelling normalized to the LemonDecider form (module filenames, auto_map, protocol and model_type); serialization weights byte-identical"
-}

LEMONDECIDER_REBRAND.json

@@ -1,16 +0,0 @@
-{
-  "schema": "lemondecider-abi-rename-receipt/1",
-  "model_name": "lemondecider-4b-merge50-20261015-v2",
-  "rebuilt_utc": "2026-10-15T07:10:31Z",
-  "purpose": "v2 spelling-normalization export of the rebranded two-route checkpoint: internal machine identifiers corrected to the LemonDecider spelling; weights and the tokenizer BPE asset byte-identical to the v1 rebranded export; v1 kept read-only with its evaluation evidence.",
-  "weights": {
-    "shards": 2,
-    "note": "serialization files unchanged in this pass"
-  },
-  "amendments": [
-    {
-      "utc": "2026-10-15T09:05:30.492782+00:00",
-      "change": "max_input_length 8192 -> 262144 in config.json and decision_config.json, enabling the Qwen3.5-4B backbone's full 256K position window (max_position_embeddings 262144, rope_theta 1e7), per user instruction 2026-10-15. Weights, tokenizer and runtime code unchanged (INDEX revision binds LEMONDECIDER_BUILD.json, untouched)."
-    }
-  ]
-}

NOTICE

@@ -1,6 +0,0 @@
-LemonDecider (LemonJev typed decision engine)
-Copyright 2026 LemonJev Project — LemonDecider
-
-Built on Qwen/Qwen3.5-4B (Apache-2.0).
-Runtime and server code adapted from Apache-2.0 and MIT upstream components
-(see LICENSE and LICENSE.code for terms).

README.md

@@ -13,9 +13,8 @@ tags:
 
 # LemonDecider 4B (LemonJev typed decision engine)
 
-**Model ID:** `lemondecider-4b-merge50-20261015-v2`
+**Model ID:** `lemondecider-4b-20261015-v2`
 **Base model:** `Qwen/Qwen3.5-4B`
-**Release date:** 2026-10-15
 
 LemonDecider 4B is a typed decision engine: it reads a state and a question with a
 list of options, runs **one forward pass**, and returns a probability distribution
@@ -73,9 +72,6 @@ label distributions.
 
 ## Package artifacts
 
-The release directory ships as a single self-contained package — weights,
-tokenizer, runtime code and provenance receipts travel together.
-
 ### Weight files
 
 | File | Size | SHA-256 |
@@ -85,8 +81,6 @@ tokenizer, runtime code and provenance receipts travel together.
 | `tokenizer.json` | 19,989,492 | `6f32ce20dc35f57a7f9ad1eac03525bd7d30f9df8cea6507e958279cc3657706` |
 | `model.safetensors.index.json` | 68,426 | `bda09c55c0152d7a6fb2db6898aea70e728dc528df6f9c026cefb0bd5d1d673b` |
 
-Provenance receipts ship inside the package: `LEMONDECIDER_BUILD.json` (build
-receipt) and `LEMONDECIDER_REBRAND.json` (export lineage).
 
 ## Evaluation results
 
@@ -228,16 +222,18 @@ Evidence: `lemondeeder-eval-20261015/jevbench/20261015-lemondecider-4b-jevbench-
 ## Usage
 
 Dependencies: `transformers==5.17.0` (needs the Qwen3.5 hybrid classes),
-`flash-linear-attention==0.5.2`, `pillow`, `torch` with SDPA. ROCm fast path
-keywords: `LEMONDECIDER_FAST`, `LEMONDECIDER_GRAPH_TOKENS`, `LEMONDECIDER_FUSED`,
-`LEMONDECIDER_CUDNN_SDPA`.
+`flash-linear-attention==0.5.2`, `pillow`, `torch` with SDPA.
+This package honours a single runtime env var: `LEMONDECIDER_CUDNN_SDPA=1`
+(keep the cuDNN SDPA backend on CUDA; defaults are kept on CPU/ROCm).
 
 ```python
-import torch
+import sys, torch
 from transformers import AutoModel
+
+ckpt = "lemondecider-4b-20261015-v2"      # local package directory
+sys.path.insert(0, ckpt)                  # make the in-package runtime importable
 from decision_utils import load_processor, predict_batch
 
-ckpt = "lemondecider-4b-merge50-20261015-v2"      # local package directory
 model = AutoModel.from_pretrained(
     ckpt, trust_remote_code=True, local_files_only=True,
     dtype=torch.bfloat16, device_map="cuda:0", attn_implementation="sdpa",
@@ -258,9 +254,11 @@ print(answers["answers"][0])   # choice, probabilities over the 3 options, confi
 
 CLI from inside the package: `python3 model.py --checkpoint . --device cuda:0 --input rows.json`.
 
-Batching: the internal batch is 8 questions per forward pass; inputs are left-padded
-to the longest prompt; requests over the 262,144-token limit are rejected with the
-limit named (HTTP 422 in server wrappers) — nothing is truncated.
+Batching: each `predict_batch` call runs ONE forward pass over all the rows you pass
+(inputs are left-padded to the longest prompt); the HTTP serving bridge chunks its
+requests at 8 questions per forward pass. Requests over the 262,144-token
+`max_input_length` limit are rejected with the limit named (HTTP 422 in server
+wrappers) — nothing is truncated.
 
 ## Intended use & limitations
 
@@ -280,15 +278,10 @@ limit named (HTTP 422 in server wrappers) — nothing is truncated.
 - `LICENSE` — Apache-2.0 text as preserved from the parent checkpoint lineage
   (base: Qwen/Qwen3.5-4B, inherit the base model's terms).
 - `LICENSE.code` — MIT for runtime/server code components adapted from upstream.
-- `NOTICE` — rebranded for this export; no new license grant is asserted.
 - No benchmark score is claimed by this package beyond the reproducible evidence
   files listed above; the Decision Index owner reserves 80% of its Full score for
   private tests, so the 50.31 public index is not a board submission.
 
-## Version history
-
-- **v1** — `lemondecider-4b-merge50-20261015-v1`: rebranded export of the merged two-route checkpoint (all evaluations in this card ran on v1; weight files byte-identical to v2).
-- **v2** — `lemondecider-4b-merge50-20261015-v2`: internal machine identifiers spelling-normalized — module files (`modeling_lemondecider.py`, `configuration_lemondecider.py`), `auto_map`, protocol `lemondeeder-v1` → `lemondecider-v1`, `model_type: lemondecider_decision`; weights and tokenizer byte-identical to v1 (sha256 unchanged).
 
 ## Acknowledgements