Remove em-dashes; replace bibtex with 'Paper coming soon'; drop Reproducibility/Intended-use sections

README.md

@@ -25,12 +25,9 @@ library_name: transformers
 
 ![FogGen overview: (a) self-aware routing at inference, (b) self-evolving training loop](./foggen_overview.png)
 
-**At a glance.** FogGen is a small, self-aware edge model that knows when to answer locally and when to defer to a stronger cloud model. The figure above summarizes the two halves of the recipe:
+FogGen is a small, self-aware edge model that knows when to answer locally and when to defer to a stronger cloud model. At inference (figure (a)) it emits a confidence score then an answer in one forward pass; if confidence `c ≥ τ` the local answer is returned, otherwise the query is routed to the cloud. Training (figure (b)) is a self-evolving loop: each round, the current checkpoint self-samples N=8 generations per question to derive confidence buckets, then SFTs on `(question, confidence, answer)` triples.
 
-- **(a) Inference — self-aware routing.** The edge model `M_N` (Qwen3-0.6B) processes a query and emits two output spans in one forward pass: a *confidence span* (positions 1–8, e.g. `Confidence: 0.75`) followed by an *answer span* (positions 10–13, e.g. `Final answer: B`). The routing decision compares the parsed confidence `c` to a threshold `τ`: if `c ≥ τ` the edge answer is returned; otherwise the query is routed to the cloud model.
-- **(b) Training — self-evolving data loop.** Each round consumes a cloud-labeled dataset (Stage 1), uses the current checkpoint `M_N` to self-sample 8 generations per question at T=0.7 to derive a confidence bucket via the `k correct → bucket` mapping (Stage 2), then trains the next checkpoint `M_{N+1}` on the resulting `(question, confidence, answer)` triples via SFT with LoRA merge (Stage 3).
-
-The released checkpoint is the endpoint (`R14`) of a 14-round continual-learning chain that trained the model across seven domains: finance, science, coding, law, math, Kazakh culture, and medicine.
+The released checkpoint is the endpoint (`R14`) of a 14-round chain trained across seven domains: finance, science, coding, law, math, Kazakh culture, medical.
 
 ## Quick demo
 
@@ -142,37 +139,11 @@ The MCQ-trained chain transfers to open-ended task types zero-shot. Local accura
 | [TruthfulQA gen](https://huggingface.co/datasets/truthfulqa/truthful_qa) | adversarial factual | 36.5% | −0.7 (anti-calibrated) |
 | [GSM8K](https://huggingface.co/datasets/openai/gsm8k) (CoT) | math word-problems | 52.0% | +2.2 |
 
-One additional round of OE training (R15, 1876 SFT rows) lifts local accuracy on these three benchmarks to 86.5% / 40.0% / 58.0% respectively — see [`issai/foggen-r15-oe`](https://huggingface.co/issai/foggen-r15-oe).
-
-## Intended use
-
-- **Edge–cloud routing systems** where a 0.6B model handles routine queries and a stronger cloud model handles harder ones.
-- **Single-pass inference**: the model produces both the answer and the routing signal in one forward pass — no separate verification calls needed.
-- **Calibrated thresholding**: change τ (default 0.5) to trade off local-vs-cloud cost.
-
-### Out-of-scope use
-
-- Standalone production use without a cloud fallback. The model is small and the confidence span is calibrated to signal *when to escalate*, not to produce gold answers on hard queries.
-- Generation tasks beyond what was tested (extractive QA, factual recall, CoT math) without additional task-type training.
-- Reliance on the confidence signal for adversarial-factuality benchmarks like TruthfulQA, where verbalized confidence is anti-calibrated by design of the dataset (see Tian et al., 2023).
-
-## Reproducibility
-
-- Per-question eval outputs and SFT inputs are released at [`issai/foggen-data`](https://huggingface.co/datasets/issai/foggen-data).
-- The 14-round chain uses LoRA SFT only (no RL), with identical hyperparameters per round.
-- Training framework: [ms-swift](https://github.com/modelscope/ms-swift) with vLLM serving for self-labeling and evaluation.
+One additional round of OE training (R15, 1876 SFT rows) lifts local accuracy on these three benchmarks to 86.5% / 40.0% / 58.0% respectively; see [`issai/foggen-r15-oe`](https://huggingface.co/issai/foggen-r15-oe).
 
 ## Citation
 
-```bibtex
-@article{foggen2026,
-  title   = {FogGen: A Self-Aware Edge-Cloud LLM Router with Verbalized Confidence Tokens},
-  author  = {Vladimir Albrekht and Akylbek Maxutov and Zhankumis Sultanova and Adil Mereke and Huseyin Atakan Varol},
-  journal = {Knowledge-Based Systems},
-  year    = {2026},
-  note    = {Under review}
-}
-```
+Paper coming soon.
 
 ## Acknowledgements