| name | long-context |
| description | Extend context windows of transformer models using RoPE, YaRN, ALiBi, and position interpolation techniques. Use when processing long documents (32k-128k+ tokens), extending pre-trained models beyond original context limits, or implementing efficient positional encodings. Covers rotary embeddings, attention biases, interpolation methods, and extrapolation strategies for LLMs. |
| version | 1.0.0 |
| author | Orchestra Research |
| license | MIT |
| tags | ["Emerging Techniques","Long Context","RoPE","YaRN","ALiBi","Position Interpolation","Extended Context","Rotary Embeddings","Attention Bias","Context Extension","Positional Encoding"] |
| dependencies | ["transformers","torch","flash-attn"] |
Long Context: Extending Transformer Context Windows
When to Use This Skill
Use Long Context techniques when you need to:
- Process long documents (32k, 64k, 128k+ tokens) with transformer models
- Extend context windows of pre-trained models (LLaMA, Mistral, etc.)
- Implement efficient positional encodings (RoPE, ALiBi)
- Train models with length extrapolation capabilities
- Deploy models that handle variable-length inputs efficiently
- Fine-tune existing models for longer contexts with minimal compute
Key Techniques: RoPE (Rotary Position Embeddings), YaRN, ALiBi (Attention with Linear Biases), Position Interpolation
Papers: RoFormer (arXiv 2104.09864), YaRN (arXiv 2309.00071), ALiBi (arXiv 2108.12409), Position Interpolation (arXiv 2306.15595)
Installation
pip install transformers torch
pip install einops
pip install rotary-embedding-torch
pip install flash-attn --no-build-isolation
Quick Start
RoPE (Rotary Position Embeddings)
import torch
import torch.nn as nn
class RotaryEmbedding(nn.Module):
"""Rotary Position Embeddings (RoPE)."""
def __init__(self, dim, max_seq_len=8192, base=10000):
super().__init__()
inv_freq = 1.0 / (base ** (torch.arange(0, dim, 2).float() / dim))
self.register_buffer("inv_freq", inv_freq)
self.max_seq_len = max_seq_len
def forward(self, seq_len, device):
t = torch.arange(seq_len, device=device).type_as(self.inv_freq)
freqs = torch.outer(t, self.inv_freq)
emb = torch.cat((freqs, freqs), dim=-1)
return emb.cos(), emb.sin()
def rotate_half(x):
"""Rotate half the hidden dimensions."""
x1, x2 = x.chunk(2, dim=-1)
return torch.cat((-x2, x1), dim=-1)
def apply_rotary_pos_emb(q, k, cos, sin):
"""Apply rotary embeddings to queries and keys."""
q_embed = (q * cos) + (rotate_half(q) * sin)
k_embed = (k * cos) + (rotate_half(k) * sin)
return q_embed, k_embed
rope = RotaryEmbedding(dim=64, max_seq_len=8192)
cos, sin = rope(seq_len=2048, device='cuda')
q_rotated, k_rotated = apply_rotary_pos_emb(query, key, cos, sin)
ALiBi (Attention with Linear Biases)
def get_alibi_slopes(num_heads):
"""Get ALiBi slope values for each attention head."""
def get_slopes_power_of_2(n):
start = 2 ** (-(2 ** -(math.log2(n) - 3)))
ratio = start
return [start * (ratio ** i) for i in range(n)]
if math.log2(num_heads).is_integer():
return get_slopes_power_of_2(num_heads)
else:
closest_power = 2 ** math.floor(math.log2(num_heads))
slopes = get_slopes_power_of_2(closest_power)
extra = get_slopes_power_of_2(2 * closest_power)
slopes.extend(extra[0::2][:num_heads - closest_power])
return slopes
def create_alibi_bias(seq_len, num_heads):
"""Create ALiBi attention bias."""
context_position = torch.arange(seq_len)
memory_position = torch.arange(seq_len)
relative_position = memory_position[None, :] - context_position[:, None]
slopes = torch.tensor(get_alibi_slopes(num_heads))
alibi = slopes[:, None, None] * relative_position[None, :, :]
return alibi
num_heads = 8
seq_len = 2048
alibi_bias = create_alibi_bias(seq_len, num_heads).to('cuda')
attn_scores = attn_scores + alibi_bias
attn_weights = torch.softmax(attn_scores, dim=-1)
Position Interpolation for LLaMA
from transformers import LlamaForCausalLM, LlamaTokenizer
model = LlamaForCausalLM.from_pretrained("meta-llama/Llama-2-7b-hf")
model.config.rope_scaling = {
"type": "linear",
"factor": 16.0
}
model.config.rope_scaling = {
"type": "dynamic",
"factor": 16.0
}
Core Concepts
1. RoPE (Rotary Position Embeddings)
How it works:
- Encodes absolute position via rotation matrix
- Provides relative position dependency in attention
- Enables length extrapolation
Mathematical formulation:
q_m = (W_q * x_m) * e^(imθ)
k_n = (W_k * x_n) * e^(inθ)
where θ_j = base^(-2j/d) for j ∈ [0, d/2)
Advantages:
- Decaying inter-token dependency with distance
- Compatible with linear attention
- Better extrapolation than absolute position encodings
2. YaRN (Yet another RoPE extensioN)
Key innovation:
- NTK-aware interpolation (Neural Tangent Kernel)
- Attention temperature scaling
- Efficient context extension (10× less tokens vs baselines)
Parameters:
yarn_config = {
"scale": 16,
"original_max_position": 2048,
"extrapolation_factor": 1.0,
"attn_factor": 1.0,
"beta_fast": 32,
"beta_slow": 1,
}
Performance:
- Extends LLaMA to 128k tokens
- 2.5× less training steps than baselines
- State-of-the-art context window extension
3. ALiBi (Attention with Linear Biases)
Core idea:
- No positional embeddings added to tokens
- Apply distance penalty directly to attention scores
- Bias proportional to key-query distance
Formula:
attention_bias[i, j] = -m * |i - j|
where m = slope for each attention head
Advantages:
- 11% faster training vs sinusoidal embeddings
- 11% less memory usage
- Strong length extrapolation (train 1k, test 2k+)
- Inductive bias towards recency
4. Position Interpolation
Technique:
- Linearly down-scale position indices
- Interpolate within trained range (vs extrapolate beyond)
- Minimal fine-tuning required
Formula:
# Original: position indices [0, 1, 2, ..., L]
# Extended: position indices [0, 0.5, 1.0, ..., L/2]
# (for 2× extension)
scaled_position[i] = i / extension_factor
Results:
- LLaMA 7B-65B extended to 32k tokens
- 1000 fine-tuning steps sufficient
- 600× better stability than extrapolation
Method Comparison
| Method | Max Context | Training Needed | Memory | Extrapolation | Best For |
|---|
| RoPE | 8k-32k | Full pre-training | Moderate | Good | New models |
| YaRN | 32k-128k | Minimal (10× efficient) | Moderate | Excellent | Extending existing models |
| ALiBi | Unlimited | Full pre-training | Low (-11%) | Excellent | Training from scratch |
| Position Interpolation | 32k+ | Minimal (1k steps) | Moderate | Poor (by design) | Quick extension |
Implementation Patterns
HuggingFace Transformers Integration
from transformers import AutoModelForCausalLM, AutoConfig
config = AutoConfig.from_pretrained("mistralai/Mistral-7B-v0.1")
config.rope_scaling = {
"type": "yarn",
"factor": 8.0,
"original_max_position_embeddings": 8192,
"attention_factor": 1.0
}
model = AutoModelForCausalLM.from_config(config)
config.rope_scaling = {
"type": "linear",
"factor": 4.0
}
config.rope_scaling = {
"type": "dynamic",
"factor": 8.0
}
Custom RoPE Implementation
class LongContextAttention(nn.Module):
"""Multi-head attention with RoPE."""
def __init__(self, hidden_size, num_heads, max_seq_len=32768):
super().__init__()
self.num_heads = num_heads
self.head_dim = hidden_size // num_heads
self.q_proj = nn.Linear(hidden_size, hidden_size)
self.k_proj = nn.Linear(hidden_size, hidden_size)
self.v_proj = nn.Linear(hidden_size, hidden_size)
self.o_proj = nn.Linear(hidden_size, hidden_size)
self.rotary_emb = RotaryEmbedding(
dim=self.head_dim,
max_seq_len=max_seq_len
)
def forward(self, hidden_states):
batch_size, seq_len, _ = hidden_states.shape
q = self.q_proj(hidden_states)
k = self.k_proj(hidden_states)
v = self.v_proj(hidden_states)
q = q.view(batch_size, seq_len, self.num_heads, self.head_dim).transpose(1, 2)
k = k.view(batch_size, seq_len, self.num_heads, self.head_dim).transpose(1, 2)
v = v.view(batch_size, seq_len, self.num_heads, self.head_dim).transpose(1, 2)
cos, sin = self.rotary_emb(seq_len, device=hidden_states.device)
q, k = apply_rotary_pos_emb(q, k, cos, sin)
attn_output = F.scaled_dot_product_attention(q, k, v)
attn_output = attn_output.transpose(1, 2).contiguous()
attn_output = attn_output.view(batch_size, seq_len, -1)
output = self.o_proj(attn_output)
return output
Fine-tuning for Long Context
Minimal Fine-tuning (Position Interpolation)
from transformers import Trainer, TrainingArguments
model.config.max_position_embeddings = 32768
model.config.rope_scaling = {"type": "linear", "factor": 16.0}
training_args = TrainingArguments(
output_dir="./llama-32k",
num_train_epochs=1,
max_steps=1000,
per_device_train_batch_size=1,
gradient_accumulation_steps=16,
learning_rate=2e-5,
warmup_steps=100,
logging_steps=10,
save_steps=500,
)
trainer = Trainer(
model=model,
args=training_args,
train_dataset=long_document_dataset,
)
trainer.train()
YaRN Fine-tuning
git clone https://github.com/jquesnelle/yarn
cd yarn
python scripts/train.py \
--model meta-llama/Llama-2-7b-hf \
--scale 16 \
--rope_theta 10000 \
--max_length 32768 \
--batch_size 1 \
--gradient_accumulation 16 \
--steps 400 \
--learning_rate 2e-5
Best Practices
1. Choose the Right Method
use_method = "ALiBi"
use_method = "YaRN"
use_method = "Position Interpolation"
use_method = "Linear RoPE Scaling"
2. Scaling Factor Selection
scaling_factor = 2.0
scaling_factor = 4.0
scaling_factor = 8.0
scaling_factor = 16.0
steps_needed = 100 * scaling_factor
3. Fine-tuning Data
train_data = [
{"text": long_doc_32k_tokens},
{"text": long_doc_24k_tokens},
{"text": long_doc_16k_tokens},
]
train_data = [
{"text": short_doc_2k_tokens},
]
4. Avoid Common Pitfalls
model.config.rope_scaling = {"type": "linear", "factor": 16.0}
model.config.rope_scaling = {"type": "linear", "factor": 16.0}
fine_tune(model, long_documents, steps=1000)
scale_to_1M_tokens()
Production Deployment
Inference with Long Context
from transformers import AutoModelForCausalLM, AutoTokenizer
model = AutoModelForCausalLM.from_pretrained(
"togethercomputer/LLaMA-2-7B-32K",
torch_dtype=torch.float16,
device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained("togethercomputer/LLaMA-2-7B-32K")
long_text = "..." * 30000
inputs = tokenizer(long_text, return_tensors="pt", truncation=False).to('cuda')
outputs = model.generate(
**inputs,
max_new_tokens=512,
temperature=0.7,
)
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
Memory Optimization
model.gradient_checkpointing_enable()
model = AutoModelForCausalLM.from_pretrained(
"meta-llama/Llama-2-7b-hf",
attn_implementation="flash_attention_2",
torch_dtype=torch.float16
)
from vllm import LLM
llm = LLM(
model="togethercomputer/LLaMA-2-7B-32K",
max_model_len=32768,
gpu_memory_utilization=0.9
)
Resources
See Also
references/rope.md - Detailed RoPE implementation and theory
references/extension_methods.md - YaRN, ALiBi, Position Interpolation comparisons
references/fine_tuning.md - Complete fine-tuning guide for context extension