Update to the <think></think> prompt used by the current code-RL runs

README.md

@@ -35,21 +35,46 @@ split stay reportable after training on this data.
 | `test` | 378 | exactly the problems in [`evalplus/mbppplus`](https://huggingface.co/datasets/evalplus/mbppplus) |
 | `heldout_mbpp_test` | 276 | MBPP's canonical test split (task_id 11-510) that is *not* in MBPP+ |
 
+## Prompt
+
+Each row is a two-turn chat asking for reasoning inside an explicit
+`<think></think>` region, then a fenced Python block:
+
+```
+system: Please reason step by step inside a <think> </think> block, then put
+        your final solution in a single ```python code block.
+user:   <task>
+        Your implementation should pass tests like the following ...
+        First reason step by step inside <think> and </think>. Then write your
+        complete solution in a single markdown fenced code block ...
+```
+
+The delimiter is deliberate. Without it, reasoning is unmarked prose and the
+only signal about it is total response length, which conflates "stopped
+reasoning" with "wrote shorter code". Measured on Qwen2.5-3B during RL on this
+data, mean response length fell from ~446 tokens at step 9 to 54 by step 95
+while training reward held at 0.83: MBPP is easy enough that reasoning earns no
+extra reward, so GRPO strips it. The tag does not prevent that collapse, but it
+makes the think region separately measurable.
+
+Half the tests (`floor(N/2)`) are shown so the model can infer the function name
+and signature; **all** tests are used for reward.
+
 ## Why 320 and not 974
 
 MBPP `full` is 974 problems across four canonical splits (prompt 10, test 500,
 validation 90, train 374). Two things are removed from the training pool:
 
-1. **Everything in MBPP+** (378 problems). MBPP+ is derived from MBPP-sanitized
-   and consumes 378 of that config's 427 problems.
+1. **Everything in MBPP+** (378). MBPP+ is derived from MBPP-sanitized and
+   consumes 378 of that config's 427 problems.
 2. **MBPP's canonical test split.** Deduplicating against MBPP+ alone would
-   leave 596 training problems, but 276 of those are canonical MBPP test — fine
+   leave 596 training problems, but 276 of those are canonical MBPP test -- fine
    if MBPP+ is your only benchmark, fatal if you ever want standard MBPP numbers.
 
 Deduplicating MBPP-**sanitized** against MBPP+ leaves only **49** problems,
 which is why the sanitized config alone is not viable for training. Each problem
-here uses its *sanitized* text and corrected tests where one exists (427 of 974
-do) while keeping the other 547 — sanitized quality at usable size.
+here uses its *sanitized* text and corrected tests where one exists (427 of 974),
+keeping the other 547 -- sanitized quality at usable size.
 
 Leakage is asserted at build time: no train `task_id` appears in MBPP+, and no
 train row comes from a withheld split.
@@ -70,17 +95,6 @@ routing. Setup imports (`test_setup_code` / `test_imports`) are folded into each
 assert string, because every test runs as a standalone script:
 `solution + "\n" + assert_case[i]`.
 
-The prompt is a two-turn chat whose system message matches the math/kk runs of
-the same study, so prompt format is not a confound across tasks:
-
-```
-system: Please reason step by step and put your final solution in a single ```python code block.
-user:   <task> + half the tests as an interface hint + fenced-block instruction
-```
-
-Half the tests (`floor(N/2)`) are shown so the model can infer the function name
-and signature; **all** tests are used for reward.
-
 ## Validation
 
 All 974 MBPP reference solutions score 1.0 through the verifier.

data/heldout_mbpp_test.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8749b790e12913372c5a3bdbc9701a4c12bd431ab9f6ed5b5fbe844e75e39406
-size 62264
+oid sha256:0e1836eb713172bc78ff15e36e26136e7f8e9e44099a42ccf2043e0b94047f10
+size 64073

data/test.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7e5b78d660746c69674a345cececc812459e6f381ce9de3683702d30c61a3e6e
-size 82958
+oid sha256:3bbe3e83af727923879172673ba789385d16fca9eae8c6fe9b49eb7777a0e399
+size 84552

data/train.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a227e2853c9ee12e0367e875dd436b3b7491663b060e034a3015cbd9d957df6d
-size 73303
+oid sha256:81b5afa5eb84b5f4b90c56ebef21cf47adb1adac68336712a77c214c9f3e2f42
+size 75058