abl_1_1_backloaded

CRUMB abl_1_1_backloaded

Model Overview

abl_1_1_backloaded is a hybrid decoder-only language model from the CRUMB (Compact Recurrent-hybrid Underlying Mamba Blocks) project. It interleaves Mamba-3 selective state-space layers with GQA (Grouped-Query Attention) layers at a 1:1 ratio with backloaded placement, and was pre-trained exclusively on Python source code.

It is one of eleven ablation variants trained to study how the Mamba-to-attention ratio and the placement of attention layers affect small-scale (~150M-parameter) language models on Python program synthesis.

Architecture

Mamba : Attention ratio — 1 : 1

6 Mamba layers + 6 GQA attention layers.

Placement — Backloaded

Attention layers are stacked at the end of the network. Layer order: M M M M M M A A A A A A

This placement keeps the final layers — closest to the output projection — precise and attention-based, while the earlier layers compress the input through linear-time Mamba blocks.

Training

Evaluation Method

Perplexity (primary metric)

Per-token cross-entropy loss with BF16 autocast, computed over the full held-out evaluation set.

Generation-based metrics

• Python syntax validity — 200 free-form completions generated per model from 49 diverse Python prompts at temperature=0.8, top_k=50, max_new_tokens=128; each completion checked with ast.parse(). Implementation: src/evaluation/syntax_validity.py.
• Qualitative side-by-side completions — 10 fixed prompts at temperature=0.6, top_k=50, max_new_tokens=200, identical random seed per prompt. Implementation: src/evaluation/qualitative_comparison.py.

Evaluation Results

Rank Summary

Out of 11 ablation configurations evaluated at the same token budget:

Within the 1:1 Mamba:Attention ratio, backloaded placement is the best of the three arrangements (5th overall). Within the backloaded arrangement, 1:1 is the weakest of the three ratios.

Intended Use & Limitations

• Domain: Python source-code language modelling.
• Base model only: no instruction tuning, no chat alignment, no safety filtering. Outputs are unconstrained code completions.
• Repetitive degeneration: all base CRUMB models tend to repeat function signatures / docstrings during free-form generation; this is expected behaviour for unaligned base models and is not specific to this variant.

Citation / Context

This model is part of the CRUMB Phase-1 ablation study:

Efficient Architectural Hybrids for Small-Scale Language Models in Python Program Synthesis — Department of Computer Science and Engineering, Daffodil International University. Findings documented in documents/phase1_ablation_findings.md.

How to Load

from tokenizers import Tokenizer
import torch
from src.model.config import CRUMBConfig
from src.model.model import CRUMBModel

config = CRUMBConfig.from_yaml("configs/model/abl_1_1_backloaded.yaml")
model = CRUMBModel(config)
state = torch.load("saved/model/abl_1_1_backloaded/model.pt", map_location="cpu")
model.load_state_dict(state)
model.eval()

tok = Tokenizer.from_file("saved/tokenizer/crumb_tok_hf/tokenizer.json")
ids = tok.encode("def fibonacci(n):\n").ids
x = torch.tensor([ids])
with torch.no_grad():
    y = model(x)