delete chat template

tokenizer: expose .vocab property for fast-tokenizer-style callers (#4)

Update README.md

Upload tech-report.pdf

Update README.md

Update README.md

fix: correct bos/eos/pad token IDs in generation_config (eos=pad=151643, bos=null)

fix tata and syn

Update README.md

Upload carbon-3b-banner.png

Update README.md

Update README.md

Upload pareto.png

Update README.md

Upload figures/inference.png

Update README.md

Add FNS code

Update README.md

model card: fix YaRN context 384 kbp → 393 kbp (65,536 tokens × 6 bp)

config: restore full config as LlamaForCausalLM (no auto_map)

config: add model_type=llama for AutoModelForCausalLM without trust_remote_code

docs: fix dtype, remove trust_remote_code for model, clean up internal comments

revert: restore LlamaForCausalLM in config.json

revert: remove auto_map, restore LlamaForCausalLM (no trust_remote_code needed)

Update README.md

tokenizer: fix decode() to handle torch tensor input via .tolist()

modeling_carbon: replace _sample() override with stable LogitsProcessor API

Update tokenizer.py

Update README.md

Update README.md

Update config.json

Update config.json

Upload modeling_carbon.py

model card: fix skip_special_tokens decode (bug fixed), remove wrong left-padding TODO

Update model card namespace references (#3)

Fix tokenizer: EOS bug + decode skip_special_tokens=True empty string (#1)

Update README.md (#2)

Update README.md

Create README.md

Initial release · sourced from hf-carbon/carbon-3B-longctx-32k-from-mix2decay-429k @ step-453000

initial commit