Rename VibeThinker-trained to VibeThinker-webAI-trained

README.md

@@ -35,9 +35,9 @@ gate, strict-7 and six-lane average of any arm in the tables below for which eac
 computed, including Qwen3-8B, Qwen3.5-4B, VibeThinker-3B and gpt-oss-120b (the 120B has no gate or strict-7
 value).
 
-![TwIL-LM3-Pro formal and general reasoning benchmarks against VibeThinker-3B, VibeThinker-trained, Qwen3.5-4B, Qwen3-8B, gpt-oss-120b, LFM2.5-8B-A1B and TwIL-LM3](benchmarks.png)
+![TwIL-LM3-Pro formal and general reasoning benchmarks against VibeThinker-3B, VibeThinker-webAI-trained, Qwen3.5-4B, Qwen3-8B, gpt-oss-120b, LFM2.5-8B-A1B and TwIL-LM3](benchmarks.png)
 
-*VibeThinker-trained is the public VibeThinker-3B after the same post-training pipeline (SLERP, d = 0.5, t = 0.5); see [Against the tuned VibeThinker-3B](#against-the-tuned-vibethinker-3b). It has no strict-7 or six-lane-average value, so those bars show n/a.*
+*VibeThinker-webAI-trained is the public VibeThinker-3B after the same post-training pipeline (SLERP, d = 0.5, t = 0.5); see [Against the tuned VibeThinker-3B](#against-the-tuned-vibethinker-3b). It has no strict-7 or six-lane-average value, so those bars show n/a.*
 
 ## Highlights
 
@@ -66,7 +66,7 @@ value).
   comes entirely from BBH-logic (0.9540 against 0.6107); without that row TwIL-LM3-Pro trails.
 * **The pipeline is not tied to one model.** The same recipe was run on five base models. On
   VibeThinker-3B it lifts the Track A macro gate from 0.374 to 0.508 (SLERP, the
-  *VibeThinker-trained* column) while the Track B 10-dataset macro moves from 0.815 to 0.802 — see
+  *VibeThinker-webAI-trained* column) while the Track B 10-dataset macro moves from 0.815 to 0.802 — see
   [One pipeline, several models](#one-pipeline-several-models).
 * **Structured formal output.** Tuned for the objects rather than the prose: FOL translation,
   entailment labels, semantic parses, Lean statements and Lean proof critique.
@@ -115,7 +115,7 @@ dedicated decode-throughput protocol: `ans/s` is defined throughout as
 `tok/s ÷ mean generation length`, so it measures completed answers rather than raw decode rate.
 Cells marked † need the engine note below.
 
-| lane / metric | TwIL-LM3-Pro | Granite-4.2-3B base | VibeThinker-3B | VibeThinker-trained ★ | Qwen3.5-4B | TwIL-LM3 | Llama-3.2-3B | LFM2-2.6B | LFM2.5-8B-A1B | Qwen3-8B | gpt-oss-120b ‡ |
+| lane / metric | TwIL-LM3-Pro | Granite-4.2-3B base | VibeThinker-3B | VibeThinker-webAI-trained ★ | Qwen3.5-4B | TwIL-LM3 | Llama-3.2-3B | LFM2-2.6B | LFM2.5-8B-A1B | Qwen3-8B | gpt-oss-120b ‡ |
 |---|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:|
 | lean_formalize token_f1 | 0.5092 | 0.2943 | 0.2087 | 0.527 | 0.4996 | 0.5869 | 0.3690 | 0.1321 | 0.4655 | 0.4022 | **0.6306** |
 | rule_induction derivation | 0.4195 | 0.2267 | 0.2038 | 0.227 | 0.5078 | 0.3192 | 0.0825 | 0.0615 | 0.1936 | 0.3680 | **0.6518** |
@@ -142,7 +142,7 @@ average cannot be computed for it; that is what the — cells mean, not a zero.
 format and tokenizer make the corpus lanes score a different quantity. The number is reported
 for completeness but is not a comparable measurement, and is excluded from the bolding.
 
-★ **VibeThinker-trained** is the public VibeThinker-3B after the same post-training pipeline as
+★ **VibeThinker-webAI-trained** is the public VibeThinker-3B after the same post-training pipeline as
 TwIL-LM3-Pro, in its SLERP (d = 0.5, t = 0.5) configuration; it is not a checkpoint in this
 repository, and [the section below](#against-the-tuned-vibethinker-3b) explains how it differs
 from the base model. Its figures are taken from the internal family comparison tables, to three
@@ -223,7 +223,7 @@ spots in absolute terms are `procedural` (strict 0.1200, loose 0.2350) and FOL t
 
 ### Track B — held-out benchmarks
 
-| dataset | TwIL-LM3-Pro | Granite-4.2-3B base | VibeThinker-3B | VibeThinker-trained ★ | Qwen3.5-4B | TwIL-LM3 | Llama-3.2-3B | LFM2-2.6B | LFM2.5-8B-A1B | Qwen3-8B | gpt-oss-120b ‡ |
+| dataset | TwIL-LM3-Pro | Granite-4.2-3B base | VibeThinker-3B | VibeThinker-webAI-trained ★ | Qwen3.5-4B | TwIL-LM3 | Llama-3.2-3B | LFM2-2.6B | LFM2.5-8B-A1B | Qwen3-8B | gpt-oss-120b ‡ |
 |---|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:|
 | gsm8k | 0.9433 | 0.9533 | 0.9600 | 0.930 | 0.8633 | 0.8733 | 0.8300 | 0.8767 | 0.9133 | 0.9567 | **0.9767** |
 | svamp | 0.9500 | 0.9200 | 0.9367 | **0.953** | 0.8867 | 0.8500 | 0.8200 | 0.9000 | 0.9133 | 0.9400 | 0.9400 |
@@ -292,19 +292,19 @@ MuSR-team sit slightly above the 2% cap-hit threshold (3.0%, 3.0% and 2.8%).
 The tables above use the public VibeThinker-3B checkpoint. The same post-training pipeline was also
 applied to it, and two of its tuned configurations are the closest same-scale comparisons to
 TwIL-LM3-Pro: WiSE-FT (λ = 0.50), and the SLERP merge (d = 0.5, t = 0.5) that appears in the
-tables and plot as **VibeThinker-trained**. These values come from the family comparison tables
+tables and plot as **VibeThinker-webAI-trained**. These values come from the family comparison tables
 rather than from a per-lane raw report, so they are shown as a summary only:
 
 | model | macro gate | macro_primary | B10 | B14 | Track A truncation |
 |---|---:|---:|---:|---:|---:|
 | TwIL-LM3-Pro | **0.554** | **0.588** | 0.790 | **0.743** | 24.2% |
-| VibeThinker-trained (SLERP, d = 0.5, t = 0.5) | 0.508 | 0.579 | **0.802** | 0.728 | — |
+| VibeThinker-webAI-trained (SLERP, d = 0.5, t = 0.5) | 0.508 | 0.579 | **0.802** | 0.728 | — |
 | VibeThinker-3B, WiSE-FT λ = 0.50 | 0.541 | **0.588** | **0.802** | 0.728 ◊ | 14.3% |
 
 ◊ There is no B14 row for the λ = 0.50 configuration; the figure is the one recorded for the SLERP
 configuration in the row above. Truncation was not recorded for the SLERP configuration.
 
-Against VibeThinker-trained, TwIL-LM3-Pro is ahead on Track A (macro gate 0.554 against 0.508,
+Against VibeThinker-webAI-trained, TwIL-LM3-Pro is ahead on Track A (macro gate 0.554 against 0.508,
 `macro_primary` 0.588 against 0.579) and on the 14-dataset macro (0.743 against 0.728), and behind
 on the 10-dataset macro (0.790 against 0.802). Against the WiSE-FT λ = 0.50 configuration the two
 are effectively tied on Track A (gate 0.554 against 0.541, `macro_primary` equal at 0.588, both
@@ -312,10 +312,10 @@ within sampling noise at n = 200), and the tuned VibeThinker-3B is ahead on the
 with a lower truncation rate. TwIL-LM3-Pro's edge is the 14-dataset macro, a gap that cannot be
 broken down per dataset from the summary values.
 
-#### How VibeThinker-trained differs from the base VibeThinker-3B
+#### How VibeThinker-webAI-trained differs from the base VibeThinker-3B
 
 **Base VibeThinker-3B** is WeiboAI's public checkpoint, unmodified, and is what the untuned columns
-in the tables above measure. **VibeThinker-trained** starts from those same weights and changes
+in the tables above measure. **VibeThinker-webAI-trained** starts from those same weights and changes
 them in two ways:
 
 * **Formal-logic post-training.** A rank-64 LoRA is trained on the same synthetic formal-logic
@@ -328,13 +328,13 @@ them in two ways:
   held-out capability in TwIL-LM3-Pro.
 
 The family tables record no reinforcement-learning (MGPO) run for VibeThinker-3B, so
-VibeThinker-trained reflects the supervised and merging stages only, whereas TwIL-LM3-Pro also has
+VibeThinker-webAI-trained reflects the supervised and merging stages only, whereas TwIL-LM3-Pro also has
 the MGPO stage. It is a reference point for the pipeline, not a checkpoint shipped in this
 repository.
 
 What that changes, on the family tables (one source, so the comparison is like for like):
 
-| metric | VibeThinker-3B (base) | VibeThinker-trained | change |
+| metric | VibeThinker-3B (base) | VibeThinker-webAI-trained | change |
 |---|---:|---:|---:|
 | macro gate | 0.374 | 0.508 | +0.134 |
 | macro_primary | 0.444 | 0.579 | +0.135 |
@@ -509,7 +509,7 @@ card, which describes the same harness. For Track B, the arms checked (including
 and Qwen3.5-4B, on all 18 tasks) share the same sampled rows and decoding, but the serving engine differs between
 arms (vLLM 0.19.1 for TwIL-LM3-Pro, its base, Qwen3.5-4B, Qwen3-8B and LFM2.5-8B-A1B; vLLM 0.11.2 for
 TwIL-LM3, Llama-3.2-3B and VibeThinker-3B), and the engine version is part of the protocol hash.
-The VibeThinker-trained column (★) comes from the internal family comparison tables and is not
+The VibeThinker-webAI-trained column (★) comes from the internal family comparison tables and is not
 covered by the manifest checks described here. Throughput has its own, separate engine caveat (see the † note under the Track A table). With
 n = 200 per lane on Track A and n = 300 per dataset on Track B, differences of two to three points
 are within sampling noise.