Falcon-H1-3B-Instruct — LiteRT-LM

tiiuae/Falcon-H1-3B-Instruct converted to the LiteRT-LM (.litertlm) format for on-device inference with Google's LiteRT-LM runtime. Requires litert-lm ≥ 0.15. Sibling of litert-community/Falcon-H1-0.5B-Instruct and litert-community/Falcon-H1-1.5B-Instruct — same conversion, same patch.

Falcon-H1 is TII's fully-hybrid design: every one of the 32 layers runs a grouped-query attention branch and a Mamba2 selective-scan branch in parallel on the same input and sums them. Each layer therefore carries both a KV cache and constant-size conv + SSM recurrent state.

File	Recipe	Size
Falcon-H1-3B-Instruct_int8.litertlm	int8 dynamic on linears + embedding (convs and the scan stay float); fp32 activations declared for GPU	3.15 GB

Correctness

• Logits parity vs PyTorch: the float export matches the HF model teacher-forced across 48 decode positions — top-1 and top-5 identical at every position, mean per-position logit correlation 1.0000, mean KL ≈ 0.
• 8-question sanity gate: 8/8 on every lane — GPU and CPU, litert-lm 0.15.0 and 0.16.0. No degeneration, no greedy flips (the first Falcon-H1 size where int8 drops nothing).
• Prompt-length robustness: hermetic prefill-chunk sweep (fresh engine per length) — CPU fills 12–51 and GPU fills 12–31 all clean.
• iPhone 17 Pro (Metal): the 8-item composite quality probe answers 8/8 on GPU and 8/8 on CPU, identical answers on both backends.

Usage

litert-lm run ./Falcon-H1-3B-Instruct_int8.litertlm --prompt "What is the capital of France? Answer in one word."

# GPU
litert-lm run ./Falcon-H1-3B-Instruct_int8.litertlm --backend gpu --cache no --prompt "..."

Multi-length prefill signatures (1–1024) are exported so the runtime picks tight chunks. The bundle carries the tokenizer and the stock ChatML-style Falcon-H1 chat template.

Performance

litert-lm benchmark (litert-lm 0.16.0), Apple M4 Max, -p 256 -d 256 --runs 3 --cache no, quiet machine:

Backend	Prefill (256)	Decode	TTFT
GPU	979 tok/s	65.3 tok/s	0.28 s
CPU	121 tok/s	20.9 tok/s	2.17 s

On device (cold start, single runs, 146-token composite prompt, quality harness):

Device	Backend	Prefill	Decode	TTFT	Peak memory
iPhone 17 Pro	GPU (Metal)	111.5 tok/s	14.0 tok/s	1.49 s	3.03 GB
iPhone 17 Pro	CPU	48.7 tok/s	7.8 tok/s	3.14 s	1.46 GB

Honest notes:

• GPU runs with fp32 activations (declared in the bundle) — expect a corresponding memory multiple over CPU.

Conversion notes

Converted with litert-torch plus a hybrid-cache patch (reproduction script + patch: hf-to-litertlm falcon_h1_work/):

• Composite hybrid cache layer: every layer holds KV + conv + recurrent state at ONE layer index — a cache layer class that is full-attention and Mamba2 at the same time (the runtime binds states by tensor name, so co-residency is just packaging).
• Folded selective scan: the Mamba2 scan is re-expressed as batched matmuls with chunk and head axes folded into the batch axis (all tensors rank ≤ 4, no BROADCAST_TO, no int64 index math) — this is what makes the graph fully delegable on GPU.
• Falcon-specific wiring: the µP multiplier vector (mup_vector, a non-persistent model-level buffer) and ssm_in_multiplier are preserved in the traced scan; the exporter's timestamp-index kwargs are re-injected at the attention layer (FalconH1's layer loop drops kwargs).
• Prefill-pad guard: the runtime runs partially-filled prefill chunks; pad positions are made exact identity steps for the SSM and the stored conv window is gathered at the last valid column.
• Quantization: post-hoc dynamic int8 over linears + embedding only; convs and the scan stay float.

License and changes

Distributed under the Falcon LLM License (inherited from the base model — see the license link). Changes from the original work: weights converted from safetensors bf16 to LiteRT flatbuffers and quantized as described above; tokenizer and chat template repackaged unmodified. This repository is a community conversion and is not affiliated with TII.