Implementation:Predibase Lorax Flash GPT2 Modeling

Knowledge Sources	Predibase_Lorax
Domains	Model_Architecture, Inference
Last Updated	2026-02-08 00:00 GMT

Overview

Optimized GPT-2 transformer implementation for LoRax inference serving with flash attention and LoRA adapter support.

Description

This module implements the GPT-2 architecture adapted for high-throughput inference in the LoRax serving framework. GPT-2 uses absolute positional embeddings (learned) rather than rotary embeddings, and employs pre-norm layer normalization. The main components are:

• FlashGPT2Attention -- Multi-head self-attention using the fused c_attn projection (combined QKV), flash attention for prefill and paged attention for decode. Supports attention scaling options including scale_attn_by_inverse_layer_idx and reorder_and_upcast_attn. Supports LoRA adapters on c_attn and c_proj via TensorParallelMultiAdapterLinear. Uses fan_in_fan_out weight layout consistent with GPT-2 conventions.
• GPT2MLP -- Feed-forward network with configurable activation function (from GPT2Config.activation_function), c_fc and c_proj projections with adapter support. Uses fan_in_fan_out weight layout.
• GPT2Block -- A single transformer block combining layer normalization (ln_1, ln_2), attention, and MLP with residual connections.
• FlashGPT2Model -- The full transformer model with token embeddings (wte), positional embeddings (wpe), stacked GPT2Block layers, and final layer normalization (ln_f).
• FlashGPT2ForCausalLM -- The top-level causal language model that wraps FlashGPT2Model and reuses the embedding weights (wte) as the LM head via weight tying.

Usage

Used internally by the LoRax server when serving GPT-2-based models. Loaded via the model registry when the model config type matches.

Code Reference

Source Location

• Repository: Predibase_Lorax
• File: server/lorax_server/models/custom_modeling/flash_gpt2_modeling.py
• Lines: 1-390

Signature

class FlashGPT2ForCausalLM(FlashGPT2PreTrainedModel):
    def __init__(self, prefix: str, config, weights):
        ...

    def forward(
        self,
        input_ids: torch.Tensor,
        position_ids: torch.Tensor,
        cu_seqlen_prefill: Optional[torch.Tensor],
        kv_cache: List[Tuple[torch.Tensor, torch.Tensor]],
        block_tables: torch.Tensor,
        slots: torch.Tensor,
        seqlen: Seqlen,
        max_s: int,
        adapter_data: AdapterBatchData,
        prefill_cache_indices: Optional[torch.Tensor] = None,
        lm_head_indices: Optional[torch.Tensor] = None,
        skip_lm_head: bool = False,
    ) -> Tuple[torch.Tensor, Optional[torch.Tensor]]:
        ...

Import

from lorax_server.models.custom_modeling.flash_gpt2_modeling import FlashGPT2ForCausalLM

I/O Contract

Inputs

Name	Type	Required	Description
input_ids	torch.Tensor	Yes	Token IDs for the input sequence
position_ids	torch.Tensor	Yes	Position indices for learned positional embeddings
cu_seqlen_prefill	Optional[torch.Tensor]	Yes	Cumulative sequence lengths for flash attention prefill; None during decode
kv_cache	List[Tuple[torch.Tensor, torch.Tensor]]	Yes	Key-value cache tensors for each layer
block_tables	torch.Tensor	Yes	Block tables for paged attention
slots	torch.Tensor	Yes	Slot indices for KV cache storage
seqlen	Seqlen	Yes	Sequence length information for the batch
max_s	int	Yes	Maximum sequence length in the batch
adapter_data	AdapterBatchData	Yes	LoRA adapter configuration for the batch
prefill_cache_indices	Optional[torch.Tensor]	No	Indices for selective cache prefilling
lm_head_indices	Optional[torch.Tensor]	No	Indices to select specific positions for LM head
skip_lm_head	bool	No	If True, return hidden states without applying the LM head

Outputs

Name	Type	Description
logits	torch.Tensor	Next-token logits over the vocabulary, computed via weight-tied embedding
speculative_logits	Optional[torch.Tensor]	Always None for GPT-2 (no speculative decoding support)

Usage Examples

# Internal usage within LoRax server
from lorax_server.models.custom_modeling.flash_gpt2_modeling import FlashGPT2ForCausalLM
# Model instantiated by model registry, not directly by users

Related Pages

• Environment:Predibase_Lorax_CUDA_GPU_Runtime