torch-tensor-parallelism

Torch Tensor Parallelism

Overview

This skill provides structured guidance for implementing tensor-parallel linear layers in PyTorch. Tensor parallelism splits individual layers across multiple devices/processes, enabling training of models too large for a single device. The two primary patterns are column parallelism (splitting output features) and row parallelism (splitting input features).

When to Use This Skill

Use this skill when:

• Implementing ColumnParallelLinear or RowParallelLinear classes
• Splitting weight matrices across ranks using torch.distributed
• Implementing layers that require all_gather or all_reduce communication patterns
• Debugging gradient flow or shape mismatches in distributed linear layers

Critical Workflow: Verify Before Implementing

Step 1: Analyze Test Requirements First

CRITICAL: Before writing any implementation code, thoroughly read and understand the test file.

Extract from tests:

1. Weight sharding expectations: Which dimension is split, what shapes are expected per rank
2. Bias handling: Is bias sharded, replicated, or handled specially
3. Input expectations: Does each rank receive full input or partitioned input
4. Output expectations: What shape should each rank produce
5. Gradient requirements: How gradients for weights and biases should match
6. World sizes tested: Typically 1, 2, and 4

Document findings before coding:

## Extracted Specification

### ColumnParallelLinear
- Weight shape per rank: [out_features/world_size, in_features]
- Bias shape per rank: [out_features/world_size]
- Input: full [batch, in_features]
- Output: partial [batch, out_features/world_size]
- Combined output: concatenate along last dimension

### RowParallelLinear
- Weight shape per rank: [out_features, in_features/world_size]
- Bias shape per rank: [out_features] (full on each rank)
- Input: partial [batch, in_features/world_size]
- Output: partial [batch, out_features]
- Combined output: sum across ranks, then add bias

Step 2: Trace Through with Concrete Examples

Before writing code, manually trace the computation with small numeric examples:

Example: world_size=2, in_features=4, out_features=2

Full weight W: [[a, b, c, d],
                [e, f, g, h]]  shape [2, 4]

Full input X: [x1, x2, x3, x4]  shape [batch, 4]

COLUMN PARALLEL (split output features):
  Rank 0: W_shard = [[a, b, c, d]]  shape [1, 4]
  Rank 1: W_shard = [[e, f, g, h]]  shape [1, 4]

  Rank 0 output: [x1*a + x2*b + x3*c + x4*d]  shape [batch, 1]
  Rank 1 output: [x1*e + x2*f + x3*g + x4*h]  shape [batch, 1]

  Gathered: concat → [y0, y1]  shape [batch, 2] ✓

ROW PARALLEL (split input features):
  Rank 0: W_shard = [[a, b], [e, f]]  shape [2, 2]
  Rank 1: W_shard = [[c, d], [g, h]]  shape [2, 2]

  Rank 0 input: [x1, x2]  shape [batch, 2]
  Rank 1 input: [x3, x4]  shape [batch, 2]

  Rank 0 partial: [x1*a + x2*b, x1*e + x2*f]  shape [batch, 2]
  Rank 1 partial: [x3*c + x4*d, x3*g + x4*h]  shape [batch, 2]

  All-reduce (sum): partial_0 + partial_1 = full output ✓

Key insight: Tracing reveals conceptual errors before implementation.

Step 3: Implement with File Verification

When writing files:

1. Write the complete implementation in a single operation
2. Immediately read back the file to verify it was written correctly
3. Check that no content was truncated

# After writing, verify:
# 1. File exists
# 2. All methods are complete (no truncated lines like "# Calculate output per)")
# 3. Forward method has return statement
# 4. All class definitions are closed

Step 4: Verify Without Running Tests

If the test environment is unavailable, verify through logical analysis:

1. Shape verification table:

   | Component | Expected Shape | Actual in Code |
   |-----------|---------------|----------------|
   | Weight    | [out/ws, in]  | ✓ chunk(dim=0) |
   | Bias      | [out/ws]      | ✓ zeros(out/ws)|
   | Output    | [batch, out/ws]| ✓ matmul result|
   

2. Trace a specific example through the code:

   • Substitute actual numbers
   • Follow each line of forward()
   • Verify final shape matches expected

3. Check gradient path:

   • Verify all parameters participate in forward computation
   • Ensure no conditional logic blocks gradient flow

Implementation Patterns

Column Parallel Linear

Splits weight along output dimension (dim 0):

class ColumnParallelLinear(nn.Module):
    def __init__(self, in_features, out_features, bias, master_weight):
        super().__init__()
        self.rank = dist.get_rank()
        self.world_size = dist.get_world_size()

        # Validate divisibility
        assert out_features % self.world_size == 0, \
            f"out_features ({out_features}) must be divisible by world_size ({self.world_size})"

        # Split weight along output dimension (dim 0)
        weight_chunks = torch.chunk(master_weight, self.world_size, dim=0)
        self.weight = nn.Parameter(weight_chunks[self.rank].contiguous())
        # Shape: [out_features/world_size, in_features]

        if bias:
            # Bias sharded same as output dimension
            local_out = out_features // self.world_size
            self.bias = nn.Parameter(torch.zeros(local_out, dtype=master_weight.dtype))
        else:
            self.register_parameter('bias', None)

    def forward(self, x):
        # x: [batch, in_features] - full input on all ranks
        output = torch.matmul(x, self.weight.t())
        # output: [batch, out_features/world_size]

        if self.bias is not None:
            output = output + self.bias

        return output  # Partial output, to be gathered

Row Parallel Linear

Splits weight along input dimension (dim 1):

class RowParallelLinear(nn.Module):
    def __init__(self, in_features, out_features, bias, master_weight):
        super().__init__()
        self.rank = dist.get_rank()
        self.world_size = dist.get_world_size()

        # Validate divisibility
        assert in_features % self.world_size == 0, \
            f"in_features ({in_features}) must be divisible by world_size ({self.world_size})"

        # Split weight along input dimension (dim 1)
        weight_chunks = torch.chunk(master_weight, self.world_size, dim=1)
        self.weight = nn.Parameter(weight_chunks[self.rank].contiguous())
        # Shape: [out_features, in_features/world_size]

        if bias:
            # Bias remains FULL on each rank
            self.bias = nn.Parameter(torch.zeros(out_features, dtype=master_weight.dtype))
        else:
            self.register_parameter('bias', None)

    def forward(self, x):
        # x: [batch, in_features/world_size] - partitioned input
        output = torch.matmul(x, self.weight.t())
        # output: [batch, out_features] - partial result

        # NOTE: In test context, bias is added AFTER all-reduce
        # Check test to determine exact bias handling
        if self.bias is not None:
            output = output + self.bias

        return output  # Partial result, to be reduced

Critical Pitfalls

1. Truncated File Writes

Problem: File write operations may truncate, leaving incomplete implementations.

Detection: Code ends mid-line (e.g., # Calculate output per)) or methods lack return statements.

Prevention:

• Always read back files after writing
• Verify all method definitions are complete
• Check that closing parentheses/braces match opening ones

2. Wrong Weight Splitting Dimension

Problem: Splitting along the wrong dimension causes shape mismatches.

Memory aid:

• Column parallel: Split dim 0 (output features) → each rank computes different outputs
• Row parallel: Split dim 1 (input features) → each rank processes different inputs

Verification: After chunking, print shapes and verify they match expected dimensions.

3. Bias Duplication in Row Parallel

Problem: Adding full bias on all ranks before all-reduce multiplies bias by world_size.

Solutions (choose based on test requirements):

• Add bias only on rank 0 (asymmetric, may affect gradients)
• Add bias/world_size on each rank (scales to correct value after sum)
• Add bias after all-reduce (requires test to handle this)

Critical: Check test file to determine expected behavior.

4. Gradient Flow Blocked by Conditionals

Problem: Using if rank == 0: for bias addition means only rank 0 receives bias gradients.

Detection: After backward pass, check bias.grad on all ranks - some may be None or zero.

Solution: Ensure bias participates in forward on all ranks, or verify test expects asymmetric gradients.

5. Skipping Manual Verification

Problem: Assuming implementation is correct without tracing through examples.

Solution: Always trace at least one concrete example (world_size=2) through the implementation before considering it complete.

6. Ignoring Edge Cases

Problem: Implementation fails for world_size=1 or non-divisible dimensions.

Verification checklist:

• Works with world_size=1 (no actual splitting)
• Works with world_size=2 (basic splitting)
• Works with world_size=4 (more granular splitting)
• Appropriate error message if dimensions not divisible

Verification Checklist

Before considering implementation complete:

• Read test file and documented exact requirements
• Traced computation manually with concrete example
• Verified file was written completely (no truncation)
• Weight split dimension matches parallelism type
• Bias handling matches test expectations
• All parameters receive gradients (trace backward pass)
• Edge cases handled (world_size=1, divisibility)
• If environment unavailable, performed logical verification with shape tables

Quick Reference

Shape Cheat Sheet

Where: O=out_features, I=in_features, ws=world_size, B=batch

Split Dimension Reference

Communication Pattern