ZhishanQ
/

qwen3-embedding-redundancy-detector-0.6B

@@ -34,17 +34,17 @@ final-answer accuracy or the coherence of the retained reasoning chain.
 ## What it does
 During reasoning, PUMA splits the model's chain of thought into steps
-$r_1, \dots, r_t$ and embeds each step with this model. A step is flagged as a
 **candidate exit point** when it is semantically redundant with a recent step,
-measured by cosine similarity:
 $$ s_t^{(k)} = \max_{t-k \le j < t} \cos\big(f(r_j),\, f(r_t)\big) $$
-where $f(\cdot)$ is this embedding model. PUMA flags step $t$ when
-$s_t^{(k)} > \tau_{\mathrm{sim}}$. The paper's default operating point is
-**$\tau_{\mathrm{sim}} = 0.35$** with a local window **$k = 1$** (compare against
-the immediately preceding step). A flagged step is then passed to PUMA's
-answer-level verification before generation is actually stopped.
 ## Training

 ## What it does
 During reasoning, PUMA splits the model's chain of thought into steps
+r₁, …, rₜ and embeds each step with this model. A step is flagged as a
 **candidate exit point** when it is semantically redundant with a recent step,
+measured by the maximum cosine similarity to the previous *k* steps:
 $$ s_t^{(k)} = \max_{t-k \le j < t} \cos\big(f(r_j),\, f(r_t)\big) $$
+where *f*(·) is this embedding model. PUMA flags a step when its redundancy
+score exceeds a threshold **τ_sim**. The paper's default operating point is
+**τ_sim = 0.35** with a local window **k = 1** (compare against the immediately
+preceding step). A flagged step is then passed to PUMA's answer-level
+verification before generation is actually stopped.
 ## Training