Rename Meridian-smaller to TwIL-LM3-Pro (model card and GGUF filenames)

.gitattributes

@@ -39,3 +39,8 @@ Meridian-smaller-Q5_K_M.gguf filter=lfs diff=lfs merge=lfs -text
 Meridian-smaller-Q6_K.gguf filter=lfs diff=lfs merge=lfs -text
 Meridian-smaller-Q8_0.gguf filter=lfs diff=lfs merge=lfs -text
 benchmarks.png filter=lfs diff=lfs merge=lfs -text
+TwIL-LM3-Pro-F16.gguf filter=lfs diff=lfs merge=lfs -text
+TwIL-LM3-Pro-Q4_K_M.gguf filter=lfs diff=lfs merge=lfs -text
+TwIL-LM3-Pro-Q5_K_M.gguf filter=lfs diff=lfs merge=lfs -text
+TwIL-LM3-Pro-Q6_K.gguf filter=lfs diff=lfs merge=lfs -text
+TwIL-LM3-Pro-Q8_0.gguf filter=lfs diff=lfs merge=lfs -text

README.md

@@ -6,7 +6,7 @@ pipeline_tag: text-generation
 base_model: ibm-granite/granite-4.2-3b
 license: other
 license_name: webai-non-commercial-license-ver.-1.0
-license_link: https://huggingface.co/webAI-Official/Meridian-smaller/blob/main/LICENSE.md
+license_link: https://huggingface.co/webAI-Official/TwIL-LM3-Pro/blob/main/LICENSE.md
 tags:
   - granite
   - granite-4.2
@@ -21,7 +21,7 @@ tags:
   - meridian
 ---
 
-# Meridian-smaller
+# TwIL-LM3-Pro
 
 A 3.66B reasoning model for **formal logic** tasks, built from
 [`ibm-granite/granite-4.2-3b`](https://huggingface.co/ibm-granite/granite-4.2-3b) through LoRA
@@ -36,7 +36,7 @@ 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).
 
-![Meridian-smaller formal and general reasoning benchmarks against VibeThinker-3B, 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, Qwen3.5-4B, Qwen3-8B, gpt-oss-120b, LFM2.5-8B-A1B and TwIL-LM3](benchmarks.png)
 
 ## Highlights
 
@@ -58,11 +58,11 @@ value).
   Qwen3-8B (0.8493 / 0.7591) and gpt-oss-120b (0.8689 / 0.8086). It gains on BBH-logic
   (0.9013 → 0.9540), MATH-500 (0.6567 → 0.7467) and MuSR (0.5922 → 0.6409), and gives back
   GSM-Symbolic (0.8900 → 0.8267) and ARC (0.8933 → 0.8600).
-* **Against VibeThinker-3B, a reasoning model of similar size.** Meridian-smaller leads it on all
+* **Against VibeThinker-3B, a reasoning model of similar size.** TwIL-LM3-Pro leads it on all
   six Track A lanes — macro gate 0.5539 against 0.4118, strict-7 0.2879 against 0.2021 — but not
   on Track B, where VibeThinker-3B is ahead on the 10-dataset macro (0.8097 against 0.7901) and
-  Meridian-smaller is ahead on the 14-dataset macro (0.7425 against 0.7262). That 14-dataset lead
-  comes entirely from BBH-logic (0.9540 against 0.6107); without that row Meridian-smaller trails.
+  TwIL-LM3-Pro is ahead on the 14-dataset macro (0.7425 against 0.7262). That 14-dataset lead
+  comes entirely from BBH-logic (0.9540 against 0.6107); without that row TwIL-LM3-Pro trails.
 * **Structured formal output.** Tuned for the objects rather than the prose: FOL translation,
   entailment labels, semantic parses, Lean statements and Lean proof critique.
 * **Runs anywhere.** 3.66B parameters in bf16 (6.82 GiB), with a Q4\_K\_M GGUF at 2.09 GiB for
@@ -79,7 +79,7 @@ assistant — there is no safety or preference tuning here beyond what Granite 4
 
 | Property                  | Value                                                                                          |
 | ------------------------- | ---------------------------------------------------------------------------------------------- |
-| Model ID                  | `webAI-Official/Meridian-smaller`                                                              |
+| Model ID                  | `webAI-Official/TwIL-LM3-Pro`                                                              |
 | Base model                | [`ibm-granite/granite-4.2-3b`](https://huggingface.co/ibm-granite/granite-4.2-3b)              |
 | Total parameters          | 3.66B (3,659,737,600)                                                                          |
 | Architecture              | Granite decoder-only dense transformer (`GraniteForCausalLM`); 40 layers, hidden size 2560, 40 attention heads / 8 KV heads |
@@ -102,7 +102,7 @@ here.
 
 ### Track A — in-domain formal logic
 
-Meridian-smaller, its base and VibeThinker-3B were run through the same harness, prompts, decoding
+TwIL-LM3-Pro, its base and VibeThinker-3B were run through the same harness, prompts, decoding
 settings and sampled rows described under [Evaluation protocol](#evaluation-protocol), and the
 other peer columns are the values already published for TwIL-LM3 on its card, produced by that
 same harness (see [Comparability](#limitations-and-caveats)). Throughput rows come from a
@@ -110,7 +110,7 @@ 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 | Meridian-smaller | Granite-4.2-3B base | VibeThinker-3B | 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 | 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 | 0.4996 | 0.5869 | 0.3690 | 0.1321 | 0.4655 | 0.4022 | **0.6306** |
 | rule_induction derivation | 0.4195 | 0.2267 | 0.2038 | 0.5078 | 0.3192 | 0.0825 | 0.0615 | 0.1936 | 0.3680 | **0.6518** |
@@ -137,7 +137,7 @@ 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.
 
-† **Throughput for Meridian-smaller, its base and VibeThinker-3B** was measured with the same
+† **Throughput for TwIL-LM3-Pro, its base and VibeThinker-3B** was measured with the same
 dedicated protocol and prompt file as the peer columns (128 prompts × 512 generated tokens, EOS
 ignored, greedy, `gpu_memory_utilization` 0.45, idle GPU), run on a single H200 with vLLM
 0.19.1 in a later session, whereas the other columns are figures recorded earlier on vLLM 0.11.2. Engine effects
@@ -147,7 +147,7 @@ against its published 15,880 (−4%), while VibeThinker-3B measured
 cells are therefore comparable with each other, and only approximately with the older columns.
 The Qwen3.5-4B figure comes from the earlier throughput sweep, which ran the Qwen3.5 models on the
 newer engine (vLLM 0.19.1) according to its driver script, so it belongs with the † cells.
-Meridian-smaller's two runs gave 20,454 and 21,884 tok/s (the first paid for cold
+TwIL-LM3-Pro's two runs gave 20,454 and 21,884 tok/s (the first paid for cold
 kernel caches) and the table shows their mean. `mean gen length` is measured directly under the
 shared evaluation protocol and is comparable across every column.
 
@@ -165,7 +165,7 @@ equal-weight mean of five objectives: the four bounded classification lanes (`en
 derivation score. In the gate, `mcq_answer` and `procedural` are credited as
 `max(exact_match, loose_match)`: for free-text answer lanes, a response that is correct but
 differently formatted is a formatting artefact rather than a reasoning failure. This affects the
-aggregate only — the per-lane rows above stay strict. (Meridian-smaller's loose-match MCQ is
+aggregate only — the per-lane rows above stay strict. (TwIL-LM3-Pro's loose-match MCQ is
 0.6500 against the 0.4100 strict figure shown in the lane row.)
 
 **`macro_primary`** is the same mean over the four classification lanes alone, without
@@ -177,23 +177,23 @@ aggregate only — the per-lane rows above stay strict. (Meridian-smaller's loos
 harsh — exact match on generative lanes is near zero for every arm — so it is useful for ranking
 models against each other but not as an absolute capability measure.
 
-Meridian-smaller leads all four summary rows that every arm with a computable value can be
+TwIL-LM3-Pro leads all four summary rows that every arm with a computable value can be
 compared on. The clearest margins are over the arms at its own scale and above: 0.5539 against
 0.3757 on the gate for LFM2.5-8B-A1B, and 0.2879 against 0.1971 on strict-7 for TwIL-LM3. Against
 Qwen3-8B the gate gap is only 0.020, but strict-7 is 0.2879 against 0.2093 — a difference that
-does not depend on loose-match credit — and Meridian-smaller leads on MCQ (strict 0.4100 against
+does not depend on loose-match credit — and TwIL-LM3-Pro leads on MCQ (strict 0.4100 against
 0.0000), rule induction (0.4195 against 0.3680) and both perplexity lanes.
 
 VibeThinker-3B — WeiboAI's 3B reasoning model, with 37.1% of its Track A generations truncated
-— trails Meridian-smaller on every objective lane and every summary row: macro gate 0.4118 against
+— trails TwIL-LM3-Pro on every objective lane and every summary row: macro gate 0.4118 against
 0.5539, strict-7 0.2021 against 0.2879, six-lane average 0.3291 against 0.5389. The gap is widest
 on Lean formalisation (0.2087 against 0.5092) and narrowest on entailment (0.5500 against 0.6700).
 Its corpus perplexities (16.93 and 27.03) are not comparable with the Granite-tokenizer models',
 because perplexity is per token and its vocabulary is 151,936 against 100,352.
 
 Qwen3.5-4B, a 4B reasoning model, sits between
-the small models and Meridian-smaller: macro gate 0.4466 against 0.5539, strict-7 0.1121 against
-0.2879. It is ahead of Meridian-smaller on two rows only, `rule_induction` (0.5078 against 0.4195,
+the small models and TwIL-LM3-Pro: macro gate 0.4466 against 0.5539, strict-7 0.1121 against
+0.2879. It is ahead of TwIL-LM3-Pro on two rows only, `rule_induction` (0.5078 against 0.4195,
 second in the table after gpt-oss-120b) and `math_corpus` perplexity (3.5926 against 3.6983), and
 far behind on entailment (0.2400 against 0.6700) and strict MCQ (0.0000 against 0.4100).
 
@@ -205,7 +205,7 @@ spots in absolute terms are `procedural` (strict 0.1200, loose 0.2350) and FOL t
 
 ### Track B — held-out benchmarks
 
-| dataset | Meridian-smaller | Granite-4.2-3B base | VibeThinker-3B | 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 | 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 | 0.8633 | 0.8733 | 0.8300 | 0.8767 | 0.9133 | 0.9567 | **0.9767** |
 | svamp | **0.9500** | 0.9200 | 0.9367 | 0.8867 | 0.8500 | 0.8200 | 0.9000 | 0.9133 | 0.9400 | 0.9400 |
@@ -230,13 +230,13 @@ spots in absolute terms are `procedural` (strict 0.1200, loose 0.2350) and FOL t
 ‡ MXFP4 weights, tensor-parallel 2 — quantized and multi-GPU, so not directly comparable to the
 single-GPU BF16 rows. ¶ 74% of its `rudas_ood` generations hit the length cap, so that cell is a
 truncation artefact rather than a measured score; excluding the row, its 13-dataset macro is
-0.8708. ¶¶ 93.7% of Meridian-smaller's `rudas_ood` generations also hit the length cap, so its
+0.8708. ¶¶ 93.7% of TwIL-LM3-Pro's `rudas_ood` generations also hit the length cap, so its
 cell is likewise a truncation artefact rather than a measurement of the model's ability.
 ¶¶¶ Qwen3.5-4B hits the length cap on 67.3% of IFEval, 59.0% of MATH-500 and 99.3% of `rudas_ood`
 generations, so those three cells are truncation artefacts too.
 
 Lengths marked ≈ are derived from stored generations rather than read from the run. For
-Meridian-smaller, its base and VibeThinker-3B they were re-tokenized directly with each model's
+TwIL-LM3-Pro, its base and VibeThinker-3B they were re-tokenized directly with each model's
 own tokenizer and
 averaged over the 14 datasets (MuSR counted once); the same method reproduces TwIL-LM3's measured
 482 to within 1%. The peer lengths marked ≈ use each model's characters-per-token ratio and are
@@ -245,7 +245,7 @@ the engine note under that table) and `ans/s` divides it by the mean generation
 VibeThinker-3B Track B rows come from the same external-baseline run and vLLM 0.11.2 engine as
 the other external peers.
 
-The honest summary of this table is that Meridian-smaller does not lead it. Larger models score
+The honest summary of this table is that TwIL-LM3-Pro does not lead it. Larger models score
 higher, and gpt-oss-120b leads seven of the fourteen dataset rows. Three things are worth
 extracting anyway. First, it holds its own base on the held-out suite (10-dataset macro 0.7901
 against 0.7942, a difference well inside the sampling noise at n = 300 per dataset) while gaining
@@ -254,14 +254,14 @@ over the base, driven by BBH-logic, MATH-500 and MuSR. Third, it edges LFM2.5-8B
 macros at less than half the parameters and leads the table outright on SVAMP (0.9500).
 
 VibeThinker-3B is the stronger held-out model on the 10-dataset macro (0.8097 against 0.7901). It
-is ahead of Meridian-smaller on seven datasets — GSM8K, ARC, LogicBench, StrategyQA, DROP,
+is ahead of TwIL-LM3-Pro on seven datasets — GSM8K, ARC, LogicBench, StrategyQA, DROP,
 MMLU-Redux and MATH-500 — and behind on the other seven: SVAMP, GSM-Symbolic, CSQA, MuSR, IFEval,
-BBH-logic and `rudas_ood`. Meridian-smaller's 0.0163 lead on the 14-dataset macro comes entirely
+BBH-logic and `rudas_ood`. TwIL-LM3-Pro's 0.0163 lead on the 14-dataset macro comes entirely
 from BBH-logic (0.9540 against 0.6107): on the other thirteen datasets it averages 0.7263 against
 VibeThinker-3B's 0.7350. VibeThinker-3B also writes much longer answers on Track B (about 1,789
 tokens against 792).
 
-Track B here was run with the chat template's thinking mode **disabled** for Meridian-smaller and
+Track B here was run with the chat template's thinking mode **disabled** for TwIL-LM3-Pro and
 its base (the prompt ends in an empty `<think></think>`), as it was for TwIL-LM3, whereas Track A
 uses the default thinking mode. The Track B numbers therefore describe non-reasoning behaviour;
 they are not a measure of what a thinking-mode generation would score. Some Track B cells are also
@@ -272,12 +272,12 @@ 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 the best tuned configuration (WiSE-FT, λ = 0.50) is the closest same-scale
-comparison to Meridian-smaller. These values come from the family comparison tables rather than
+comparison to TwIL-LM3-Pro. 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 |
 |---|---:|---:|---:|---:|---:|
-| Meridian-smaller | **0.554** | **0.588** | 0.790 | **0.743** | 24.2% |
+| TwIL-LM3-Pro | **0.554** | **0.588** | 0.790 | **0.743** | 24.2% |
 | 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 best tuned VibeThinker-3B
@@ -285,7 +285,7 @@ B14 in the family table (SLERP, d = 0.5, t = 0.5).
 
 The two are effectively tied on Track A (macro gate 0.554 against 0.541, `macro_primary` equal at
 0.588, both within sampling noise at n = 200), and the tuned VibeThinker-3B is ahead on the
-10-dataset macro (0.802 against 0.790) with a lower truncation rate. Meridian-smaller's edge is the
+10-dataset macro (0.802 against 0.790) with a lower truncation rate. TwIL-LM3-Pro's edge is the
 14-dataset macro (0.743 against 0.728), a gap that cannot be broken down per dataset from the
 summary values. Read gaps to the
 *untuned* VibeThinker-3B within one source: the family table records that checkpoint at a gate of
@@ -319,7 +319,7 @@ is a selection tool, not a benchmark claim.
 import torch
 from transformers import AutoModelForCausalLM, AutoTokenizer
 
-model_id = "webAI-Official/Meridian-smaller"
+model_id = "webAI-Official/TwIL-LM3-Pro"
 tok = AutoTokenizer.from_pretrained(model_id)
 model = AutoModelForCausalLM.from_pretrained(
     model_id, torch_dtype=torch.bfloat16, device_map="auto"
@@ -357,14 +357,14 @@ as EOS, so run it in conversation mode (`-cnv`).
 
 | file | quant | size | bits/weight | notes |
 |---|---|---:|---:|---|
-| `Meridian-smaller-Q4_K_M.gguf` | Q4_K_M | 2.09 GiB | 4.91 | recommended default; runs on CPU or 4 GB of VRAM |
-| `Meridian-smaller-Q5_K_M.gguf` | Q5_K_M | 2.43 GiB | 5.71 | a little more headroom than Q4_K_M |
-| `Meridian-smaller-Q6_K.gguf` | Q6_K | 2.80 GiB | 6.57 | close to Q8_0 quality at about three-quarters the size |
-| `Meridian-smaller-Q8_0.gguf` | Q8_0 | 3.63 GiB | 8.51 | near-lossless, for quality-sensitive use |
-| `Meridian-smaller-F16.gguf` | F16 | 6.82 GiB | 16.01 | unquantized, for requantization or reference runs |
+| `TwIL-LM3-Pro-Q4_K_M.gguf` | Q4_K_M | 2.09 GiB | 4.91 | recommended default; runs on CPU or 4 GB of VRAM |
+| `TwIL-LM3-Pro-Q5_K_M.gguf` | Q5_K_M | 2.43 GiB | 5.71 | a little more headroom than Q4_K_M |
+| `TwIL-LM3-Pro-Q6_K.gguf` | Q6_K | 2.80 GiB | 6.57 | close to Q8_0 quality at about three-quarters the size |
+| `TwIL-LM3-Pro-Q8_0.gguf` | Q8_0 | 3.63 GiB | 8.51 | near-lossless, for quality-sensitive use |
+| `TwIL-LM3-Pro-F16.gguf` | F16 | 6.82 GiB | 16.01 | unquantized, for requantization or reference runs |
 
 ```bash
-llama-cli -hf webAI-Official/Meridian-smaller:Q4_K_M -cnv --temp 0 -n 2048
+llama-cli -hf webAI-Official/TwIL-LM3-Pro:Q4_K_M -cnv --temp 0 -n 2048
 ```
 
 Two things matter for reproducing the scores above under llama.cpp. Pass `--temp 0`, because the
@@ -413,12 +413,12 @@ makes no claim about those. The weak absolute areas inside the specialisation ar
 (exact match 0.0100), `procedural` (strict 0.1200) and semantic parsing exact match (0.0000);
 `rule_induction` parses only 56.5% of outputs.
 
-**Comparability.** For Track A, Meridian-smaller, its base, VibeThinker-3B, Qwen3.5-4B, Qwen3-8B, LFM2-2.6B,
+**Comparability.** For Track A, TwIL-LM3-Pro, its base, VibeThinker-3B, Qwen3.5-4B, Qwen3-8B, LFM2-2.6B,
 LFM2.5-8B-A1B and Llama-3.2-3B were checked to share the same sampled-row manifest and dataset
 hash, seed and decoding; the TwIL-LM3 and gpt-oss-120b values are carried over from the TwIL-LM3
 card, which describes the same harness. For Track B, the arms checked (including VibeThinker-3B
 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 Meridian-smaller, its base, Qwen3.5-4B, Qwen3-8B and LFM2.5-8B-A1B; vLLM 0.11.2 for
+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.
 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
@@ -435,7 +435,7 @@ are within sampling noise.
   sampled rows.
 - Throughput: 128 prompts drawn from a fixed Track A prompt file, 512 generated tokens each with
   EOS ignored, greedy, vLLM `gpu_memory_utilization` 0.45, `max_model_len` 4096, on an otherwise
-  idle GPU (a single H200 for the Meridian-smaller, base and VibeThinker-3B runs); the reported rate is generated tokens over decode time, excluding engine start-up and
+  idle GPU (a single H200 for the TwIL-LM3-Pro, base and VibeThinker-3B runs); the reported rate is generated tokens over decode time, excluding engine start-up and
   compilation.
 
 `repetition_penalty = 1.0` is load-bearing. A 1.1 penalty produced apparent 20-point swings on
@@ -444,7 +444,7 @@ identity so a mismatched runner fails loudly instead of quietly producing a diff
 
 ## Relationship to TwIL-LM
 
-Meridian-smaller applies the same post-training pipeline as the
+TwIL-LM3-Pro applies the same post-training pipeline as the
 [TwIL-LM3](https://huggingface.co/webAI-Official/TwIL-LM3) and TwIL-LM family — LoRA SFT,
 checkpoint fusion, WiSE-FT and MGPO — to a different base, IBM's Granite 4.2 3B, instead of
 SmolLM3 or SmolLM2 with some additional mechanisms. Compared with TwIL-LM3 it is a stronger in-domain model (macro gate 0.5539