jdecim/PIT-4B-FT-202112-earnings-SFT

PIT-4B-FT-202112-earnings-SFT

LoRA adapter that fine-tunes `Diamegs/PIT-4B-FT-202112` on synthetic + natural earnings-call QA.

This is the 2021-12 PIT-horizon variant. The companion 2022-12 model is at `jdecim/PIT-4B-FT-202212-earnings-SFT`.

Model details

• —Base model: Diamegs/PIT-4B-FT-202112
• —Adapter type: LoRA (PEFT)
• —PIT horizon: 2021-12. Training data and base model both respect the 202112 point-in-time cutoff — no future knowledge by construction.
• —Training dataset: `jdecim/pit-earnings-call-qa` splits 202112/sft_train.jsonl / sft_val.jsonl.

Prompt format

Same <|user|> / <|assistant|> / <|end|> tags as the PIT base:

<|user|>
Context:
{context}

Question: {question}
<|assistant|>
{answer}<|end|>

Usage

from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

base = "Diamegs/PIT-4B-FT-202112"
adapter = "jdecim/PIT-4B-FT-202112-earnings-SFT"

tok = AutoTokenizer.from_pretrained(base, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(
    base, torch_dtype=torch.bfloat16, trust_remote_code=True, device_map="auto",
)
model = PeftModel.from_pretrained(model, adapter)
model.eval()

Evaluation

General NLP benchmarks

Pending. This model has not been benchmarked directly with lm-eval-harness; results will be added after evaluation.

Earnings-call benchmark (jdecim/pit-earnings-call-qa @ 202112/benchmark_1k.jsonl)

Pending. See src/evaluate_benchmark_metrics.py and runai/run_benchmark_eval.sh for the evaluation pipeline.

Limitations

• —2048-token context limit (inherited from base).
• —LoRA adapter — must be loaded on top of Diamegs/PIT-4B-FT-202112.
• —Strict 2021-12 PIT horizon; not appropriate for analyzing later quarters.

Citation

Built atop the PIT-4B-FT family from Kelly et al. (2026) "Scaling Point-in-Time Language Models".