Model card: add FullDuplexBench v1.0 + v1.5 results vs base

README.md

@@ -9,18 +9,72 @@ tags:
 
 # MiniCPM-duplex-rl
 
-RL-trained **full-duplex** conversational model tuned for natural turn-taking — the training
-goal is to **avoid interrupting the user** while still responding promptly once they finish.
+A full-duplex turn-taking RL fine-tune of
+[`enochlev/MiniCPM-duplex`](https://huggingface.co/enochlev/MiniCPM-duplex)
+(MiniCPM-duplex, from `xinrongzhang2022/MiniCPM-duplex`). The model decides every
+~1.7 s block whether to speak or stay silent while the user may also be speaking;
+this checkpoint was trained with REINFORCE over block-level turn-taking rewards
+(interruption penalties, timely-response rewards, silence penalties) for 180 steps.
 
-- **Base:** [enochlev/MiniCPM-duplex](https://huggingface.co/enochlev/MiniCPM-duplex)
-- **Method:** REINFORCE over turn-taking reward functions (interruption penalty, timely-response
-  reward, idle/silence shaping). This checkpoint is seed 2 ("run9"), best avg_reward +1.35.
-- **Serving:** text-only full-duplex server (`server.py --cpm`), ~1.7 s decision blocks,
-  monotonic ASR commit.
+**Effect:** compared to the base model it interrupts the user less, yields to
+barge-ins, and resumes after overlapping speech — trading away some take-turn
+responsiveness on direct interruptions.
 
-## Evaluation (FullDuplexBench)
+Training + serving code: [enochlev/text-only-duplex-model](https://github.com/enochlev/text-only-duplex-model)
 
-Scored with GPT-4o behavior classification over ~90–175 real conversations per task. On the
-"don't interrupt" objective (lower RESPOND / higher RESUME = interrupts less), this seed
-interrupts less than the base — and less than an independent same-config seed — on 3 of 4 tasks
-(largest gains on `background_speech` and `talking_to_other`; `user_backchannel` already at floor).
+## Serving
+
+Serve **bf16** (fp8 + greedy sampling breaks the idle/speak decision):
+
+```bash
+vllm serve enochlev/MiniCPM-duplex-rl \
+    --served-model-name cpm-text-duplex --max-model-len 3000 \
+    --gpu_memory_utilization 0.30 --trust-remote-code
+```
+
+then point the repo's `server.py --cpm` at it for the real-time audio stack
+(Kokoro TTS + Parakeet ASR + WebSocket client protocol).
+
+## FullDuplexBench results (base vs this model)
+
+Evaluated with [Full-Duplex-Bench](https://github.com/DanielLin94144/Full-Duplex-Bench)
+(GPT-4o behavior classification).
+
+### v1.5 — behavior distribution + stop/response latency (pooled, seconds)
+
+| Task (desired) | Model | n | RESPOND | RESUME | Stop (s) | Resp (s) |
+|---|---|---|---|---|---|---|
+| user_interruption (RESPOND ↑) | base | 200 | **0.65** | 0.20 | 2.17 | 1.93 |
+| | rl | 175 | 0.49 | 0.36 | 2.21 | 2.60 |
+| user_backchannel (RESUME ↑) | base | 98 | 0.00 | 0.52 | 0.73 | 1.93 |
+| | rl | 98 | 0.00 | **0.63** | 0.67 | 2.03 |
+| talking_to_other (RESUME ↑) | base | 100 | 0.47 | 0.24 | 1.43 | 1.90 |
+| | rl | 100 | 0.28 | **0.43** | 1.53 | 2.18 |
+| background_speech (RESUME ↑) | base | 100 | 0.63 | 0.25 | 1.21 | 2.27 |
+| | rl | 98 | 0.45 | **0.31** | 1.19 | 2.32 |
+
+The RL model wins the three tasks whose desired behavior is *staying quiet /
+resuming* (backchannels, third-party speech, background speech) and is less eager
+on direct user interruptions.
+
+### v1.0 — turn-taking dimensions
+
+| Metric | base | rl |
+|---|---|---|
+| Candor Pause Handling · take-turn | 0.916 | 0.635 |
+| Candor Turn Taking · take-turn / latency | 0.992 / 0.31s | 0.861 / 0.85s |
+| ICC Backchannel · JSD / TOR / Freq | 0.44 / 0.71 / 0.44 | 0.69 / 0.73 / 0.15 |
+| Synthetic Pause Handling · take-turn | 0.934 | 0.653 |
+| Synthetic User Interruption · rating / take-turn / latency | 4.15 / 1.0 / 0.71s | 4.04 / 0.98 / 1.76s |
+
+v1.0's take-turn/latency conventions favor the eager base model; the consistent
+direction across both versions reflects the RL objective — restraint over
+eagerness.
+
+## Training summary
+
+- 180 REINFORCE steps, lr 5e-6, 32 episodes/step, γ=0.90, per-batch z-scored advantages
+- Block-level rewards: interruption penalty, timely-response reward, silence
+  penalty, missed-turn penalty, backchannel-loop penalty
+- Seed-reproducible (two independent seeds: best avg reward +1.36 / +1.35);
+  replay eval cut stale-overlap speech ~45% vs base without going over-silent