Upload 5 files

README.md

@@ -0,0 +1,37 @@
+---
+license: cc-by-4.0
+tags:
+- research-notes
+- image-captioning
+---
+
+# Notes on Image Captioning
+
+## Repository summary
+
+Reading notes and an experiment sketch for **Image Captioning**. The repository emphasizes what still needs to be tested instead of manufacturing scores or release claims.
+
+## What is covered
+
+- the scope of the research question and likely confounders
+- a proposed comparison with matched baselines
+- concrete evaluation context such as MS COCO Captions, NoCaps, and TextCaps
+- reproducibility checks, failure modes, and open questions
+- topic-relevant references
+
+## How to read this repository
+
+Start with `analysis.md` for the full note. Sections labeled as plans or hypotheses should not be interpreted as experimental results. If results are added later, they should include dataset versions, commands, seeds, hardware, and raw logs.
+
+## Scope and limitations
+
+The note is intentionally exploratory. It does not claim benchmark improvements, completed ablations, released code, or a trained checkpoint. References and proposed datasets provide a starting point for verification rather than evidence that the study has already been run.
+
+## Files
+
+- `analysis.md` — primary artifact
+- `README.md` — this documentation
+
+## License
+
+Released under **cc-by-4.0**. Review the source-data terms separately when this repository is used with external datasets.

analysis.md

@@ -0,0 +1,54 @@
+# Image Captioning: Research Notes
+
+## Status
+
+Working note / experiment plan. No completed benchmark results are claimed here.
+
+## 1. Scope and motivation
+
+The goal of this note is to turn a broad idea into a falsifiable study of faithful generation that mentions visible evidence without hallucinating objects. The central question is whether the proposed change improves the target behavior under a matched training and evaluation budget. The note deliberately separates hypotheses from observations so that future results can be added without rewriting the rationale.
+
+## 2. Context
+
+Research on image captioning often mixes improvements from architecture, data scale, preprocessing, and compute. A useful comparison therefore needs controlled baselines and explicit reporting of resource use. For this topic, the main confound is that n-gram metrics only partially reflect factuality and usefulness.
+
+## 3. Working hypothesis
+
+A focused change to the representation or interaction mechanism may improve CIDEr without increasing deployment cost disproportionately. The hypothesis should be rejected if gains disappear after matching parameter count, data exposure, or tuning budget.
+
+## 4. Proposed approach
+
+The first implementation should keep modality-specific preprocessing simple, project inputs into a shared representation space, and isolate the new component behind a small interface. Baselines should include a comparable model without the component and a stronger off-the-shelf reference. Any optimization should be applied to all systems, not only the proposed one.
+
+## 5. Evaluation plan
+
+| Dataset | Role | Primary measure |
+|---|---|---|
+| MS COCO Captions | primary evaluation | CIDEr |
+| NoCaps | transfer / robustness | SPICE |
+| TextCaps | transfer / robustness | human preference |
+
+Planned comparisons include a matched-capacity baseline, an ablation that removes the proposed component, and an out-of-domain transfer check. Default training values for the first controlled run are learning rate `0.0003`, batch size `16`, and `3` independent seeds. These are planning values, not claims about a finished experiment.
+
+## 6. Reproducibility checklist
+
+- Fix preprocessing before tuning.
+- Report mean and standard deviation across seeds.
+- Keep a held-out error-analysis split.
+- Record wall-clock time and peak memory.
+
+## 7. Failure modes and responsible use
+
+The analysis should report subgroup and category-level failures instead of relying only on a single aggregate score. Particular attention is needed because n-gram metrics only partially reflect factuality and usefulness. No production use is recommended without task-specific validation, data review, and an assessment of privacy and bias.
+
+## 8. Open questions
+
+- Does the proposed component improve calibration as well as the primary metric?
+- How sensitive is the conclusion to preprocessing and random seed?
+- Can a simpler baseline recover the same gain with more careful tuning?
+
+## References
+
+[1] Chen et al., COCO Captions, 2015.
+[2] Agrawal et al., NoCaps, 2019.
+[3] Sidorov et al., TextCaps, 2020.

config.json

@@ -0,0 +1,16 @@
+{
+  "architectures": [
+    "CustomResearchModel"
+  ],
+  "architecture": "transformer",
+  "model_type": "transformer",
+  "hidden_size": 256,
+  "num_hidden_layers": 6,
+  "num_attention_heads": 4,
+  "intermediate_size": 1024,
+  "hidden_act": "gelu",
+  "max_position_embeddings": 512,
+  "layer_norm_eps": 1e-12,
+  "checkpoint_status": "initialization-only",
+  "notes": "Untrained checkpoint for smoke tests; no benchmark claim."
+}

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fce82238203c1dd954c3c2566b9161bbb3d124921e7de7ad51bec678307ab7c2
+size 132832

training_args.json

@@ -0,0 +1,9 @@
+{
+  "optimizer": "adamw",
+  "scheduler": "cosine",
+  "learning_rate": 5e-05,
+  "batch_size": 16,
+  "epochs": 10,
+  "seed": 3407,
+  "status": "default recipe; not a completed run"
+}