Close non-thinking chat prefix

README.md

@@ -81,7 +81,7 @@ optional TileLang inference implementation are distributed with this release.
 
 | Field | Value |
 | --- | --- |
-| Repository | `aabbdev/RWKV7-1.5B-20260805` |
+| Repository | `BlinkDL/RWKV7-1.5B-20260805` |
 | Architecture class | `Rwkv7ForCausalLM` |
 | Public size label | `1.5`B |
 | Source parameters | `1,527,668,736` |
@@ -116,7 +116,7 @@ from transformers import (
     PreTrainedConfig,
 )
 
-model_id = "aabbdev/RWKV7-1.5B-20260805"
+model_id = "BlinkDL/RWKV7-1.5B-20260805"
 tokenizer = AutoTokenizer.from_pretrained(
     model_id,
     config=PreTrainedConfig(),
@@ -144,7 +144,7 @@ THINK_RE = re.compile(r"\A<think>?\s*(.*?)\s*</think>?", re.DOTALL)
 
 
 def assistant_content(completion, thinking, *, close_incomplete=False):
-    prefix = "<think" if thinking else "<think></think"
+    prefix = "<think" if thinking else "<think></think>\n"
     reply = prefix + completion
     thinking_block = THINK_RE.match(reply)
     if thinking:
@@ -153,7 +153,7 @@ def assistant_content(completion, thinking, *, close_incomplete=False):
         return f"{reply.rstrip()}\n</think>".strip()
     return "" if thinking_block is None else reply[thinking_block.end():].strip()
 
-model_id = "aabbdev/RWKV7-1.5B-20260805"
+model_id = "BlinkDL/RWKV7-1.5B-20260805"
 tokenizer = AutoTokenizer.from_pretrained(
     model_id,
     config=PreTrainedConfig(),
@@ -203,10 +203,11 @@ print(
 ```
 
 Set `thinking=True` for the RWKV thinking prefix. The intentional generation
-prefixes are `Assistant: <think></think` and `Assistant: <think`; do not append a
-closing `>` to them. The post-processing above reconstructs that prefix before
-removing an empty thinking block or preserving an enabled one. If generation hits
-the token limit inside thinking, it closes the displayed block before returning it.
+prefixes are `Assistant: <think></think>` followed by a newline and
+`Assistant: <think`. Only the enabled thinking prefix intentionally leaves its opening
+tag incomplete. The post-processing above reconstructs that prefix before removing an
+empty thinking block or preserving an enabled one. If generation hits the token limit
+inside thinking, it closes the displayed block before returning it.
 Reference stops are token ID `0` and `\n\nUser:`.
 
 Strip trailing spaces from user input. The official RWKV prompt guide is available
@@ -218,14 +219,14 @@ Install the versions listed in `inference/requirements.txt`, then run the bundle
 interactive chat:
 
 ```bash
-python inference/generate.py --model aabbdev/RWKV7-1.5B-20260805 --backend auto --interactive
+python inference/generate.py --model BlinkDL/RWKV7-1.5B-20260805 --backend auto --interactive
 ```
 
 Or independent prompts separated by blank lines:
 
 ```bash
 python inference/generate.py \
-  --model aabbdev/RWKV7-1.5B-20260805 \
+  --model BlinkDL/RWKV7-1.5B-20260805 \
   --backend auto \
   --input-file prompts.txt
 ```

chat_template.jinja

@@ -51,6 +51,6 @@
 {%- if thinking -%}
 {{ 'Assistant: <think' }}
 {%- else -%}
-{{ 'Assistant: <think></think' }}
+{{ 'Assistant: <think></think>\n' }}
 {%- endif -%}
 {%- endif -%}

inference/generate.py

@@ -59,7 +59,7 @@ class StopOnText(StoppingCriteria):
 def _assistant_content(
     completion: str, thinking: bool, *, close_incomplete: bool = False
 ) -> str:
-    prefix = "<think" if thinking else "<think></think"
+    prefix = "<think" if thinking else "<think></think>\n"
     reply = prefix + completion
     thinking_block = THINK_RE.match(reply)
     if thinking:

release-manifest.json

@@ -1,7 +1,7 @@
 {
   "builder": {
     "asset_set": "2026.08.06-r1",
-    "assets_sha256": "3c3f6237fcd9dc2ec8ba0b61b626ef72924ebe42bdc6dc49e21a5bfeb19bb96b",
+    "assets_sha256": "5a31129dcca03f5ecd55157e1561af64aaf35046c77e253fc4d6214b95ab8ede",
     "version": "0.2.0"
   },
   "conversion": {
@@ -36,13 +36,13 @@
     },
     "README.md": {
       "role": "model_card",
-      "sha256": "b7795213e1fcf1d9e7de41fa027a0b2d311b2dacfa5a5ba9f62c4c2484bd729f",
-      "size_bytes": 10260
+      "sha256": "0f1a8bfe6a0d420333fdf5f7e787e751104348c81c531e3b846a6b0bf7b88d44",
+      "size_bytes": 10331
     },
     "chat_template.jinja": {
       "role": "tokenizer",
-      "sha256": "f0d8f8161f03d38d0bf68b1ea9a1293064c0fe2b60dfc2874008154c33861c07",
-      "size_bytes": 2333
+      "sha256": "abcd52a871eb3f7f0de7b69f2e0ac974fd7cfb03d4acea7a4609cb441f5d9048",
+      "size_bytes": 2336
     },
     "config.json": {
       "role": "model_config",
@@ -61,8 +61,8 @@
     },
     "inference/generate.py": {
       "role": "inference",
-      "sha256": "9e4f3461ade971a96a276ec4e18683d6d8df6420b09bf3e63cc499b6e119aa05",
-      "size_bytes": 7804
+      "sha256": "8c2ca021f04b15db07e3724c783867ffeb7e323d71f753523094b69d2338b8f0",
+      "size_bytes": 7807
     },
     "inference/kernel.py": {
       "role": "inference",
@@ -320,7 +320,7 @@
       }
     }
   },
-  "schema_version": 6,
+  "schema_version": 7,
   "source": {
     "filename": "rwkv7-g1i-1.5b-20260805-ctx16384.pth",
     "kind": "huggingface",