Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

[Issue]: Expected dout_seq_stride == out_seq_stride to be true, but got false #40

Open
ehartford opened this issue Jan 30, 2024 · 14 comments · May be fixed by #58
Open

[Issue]: Expected dout_seq_stride == out_seq_stride to be true, but got false #40

ehartford opened this issue Jan 30, 2024 · 14 comments · May be fixed by #58

Comments

@ehartford
Copy link

Problem Description

[rank4]:     _flash_attn_varlen_backward(
[rank4]:   File "/shared/prod/home/ehartford/.conda/envs/axolotl/lib/python3.11/site-packages/flash_attn-2.0.4-py3.11-linux-x86_64.egg/flash_attn/flash_attn_interface.py", line 79, in _flash_attn_varlen_backward
[rank4]:     dq, dk, dv, softmax_d, = flash_attn_cuda.varlen_bwd(
[rank4]:                              ^^^^^^^^^^^^^^^^^^^^^^^^^^^
[rank4]: RuntimeError: Expected dout_seq_stride == out_seq_stride to be true, but got false.  (Could this error message be improved?  If so, please report an enhancement request to PyTorch.)```

### Operating System

Ubuntu 22.04.1 LTS (Jammy Jellyfish)

### CPU

AMD EPYC 7763 64-Core Processor  

### GPU

AMD Instinct MI210

### ROCm Version

ROCm 5.7.1

### ROCm Component

_No response_

### Steps to Reproduce

_No response_

### (Optional for Linux users) Output of /opt/rocm/bin/rocminfo --support

(axolotl) ehartford@smc-r08-08:~/axolotl$ rocminfo --support
ROCk module is loaded
=====================    
HSA System Attributes    
=====================    
Runtime Version:         1.1
System Timestamp Freq.:  1000.000000MHz
Sig. Max Wait Duration:  18446744073709551615 (0xFFFFFFFFFFFFFFFF) (timestamp count)
Machine Model:           LARGE                              
System Endianness:       LITTLE                             
Mwaitx:                  DISABLED
DMAbuf Support:          YES

==========               
HSA Agents               
==========               
*******                  
Agent 1                  
*******                  
  Name:                    AMD EPYC 7763 64-Core Processor    
  Uuid:                    CPU-XX                             
  Marketing Name:          AMD EPYC 7763 64-Core Processor    
  Vendor Name:             CPU                                
  Feature:                 None specified                     
  Profile:                 FULL_PROFILE                       
  Float Round Mode:        NEAR                               
  Max Queue Number:        0(0x0)                             
  Queue Min Size:          0(0x0)                             
  Queue Max Size:          0(0x0)                             
  Queue Type:              MULTI                              
  Node:                    0                                  
  Device Type:             CPU                                
  Cache Info:              
    L1:                      32768(0x8000) KB                   
  Chip ID:                 0(0x0)                             
  ASIC Revision:           0(0x0)                             
  Cacheline Size:          64(0x40)                           
  Max Clock Freq. (MHz):   2450                               
  BDFID:                   0                                  
  Internal Node ID:        0                                  
  Compute Unit:            64                                 
  SIMDs per CU:            0                                  
  Shader Engines:          0                                  
  Shader Arrs. per Eng.:   0                                  
  WatchPts on Addr. Ranges:1                                  
  Features:                None
  Pool Info:               
    Pool 1                   
      Segment:                 GLOBAL; FLAGS: FINE GRAINED        
      Size:                    264004740(0xfbc6484) KB            
      Allocatable:             TRUE                               
      Alloc Granule:           4KB                                
      Alloc Alignment:         4KB                                
      Accessible by all:       TRUE                               
    Pool 2                   
      Segment:                 GLOBAL; FLAGS: KERNARG, FINE GRAINED
      Size:                    264004740(0xfbc6484) KB            
      Allocatable:             TRUE                               
      Alloc Granule:           4KB                                
      Alloc Alignment:         4KB                                
      Accessible by all:       TRUE                               
    Pool 3                   
      Segment:                 GLOBAL; FLAGS: COARSE GRAINED      
      Size:                    264004740(0xfbc6484) KB            
      Allocatable:             TRUE                               
      Alloc Granule:           4KB                                
      Alloc Alignment:         4KB                                
      Accessible by all:       TRUE                               
  ISA Info:                
*******                  
Agent 2                  
*******                  
  Name:                    AMD EPYC 7763 64-Core Processor    
  Uuid:                    CPU-XX                             
  Marketing Name:          AMD EPYC 7763 64-Core Processor    
  Vendor Name:             CPU                                
  Feature:                 None specified                     
  Profile:                 FULL_PROFILE                       
  Float Round Mode:        NEAR                               
  Max Queue Number:        0(0x0)                             
  Queue Min Size:          0(0x0)                             
  Queue Max Size:          0(0x0)                             
  Queue Type:              MULTI                              
  Node:                    1                                  
  Device Type:             CPU                                
  Cache Info:              
    L1:                      32768(0x8000) KB                   
  Chip ID:                 0(0x0)                             
  ASIC Revision:           0(0x0)                             
  Cacheline Size:          64(0x40)                           
  Max Clock Freq. (MHz):   2450                               
  BDFID:                   0                                  
  Internal Node ID:        1                                  
  Compute Unit:            64                                 
  SIMDs per CU:            0                                  
  Shader Engines:          0                                  
  Shader Arrs. per Eng.:   0                                  
  WatchPts on Addr. Ranges:1                                  
  Features:                None
  Pool Info:               
    Pool 1                   
      Segment:                 GLOBAL; FLAGS: FINE GRAINED        
      Size:                    264179616(0xfbf0fa0) KB            
      Allocatable:             TRUE                               
      Alloc Granule:           4KB                                
      Alloc Alignment:         4KB                                
      Accessible by all:       TRUE                               
    Pool 2                   
      Segment:                 GLOBAL; FLAGS: KERNARG, FINE GRAINED
      Size:                    264179616(0xfbf0fa0) KB            
      Allocatable:             TRUE                               
      Alloc Granule:           4KB                                
      Alloc Alignment:         4KB                                
      Accessible by all:       TRUE                               
    Pool 3                   
      Segment:                 GLOBAL; FLAGS: COARSE GRAINED      
      Size:                    264179616(0xfbf0fa0) KB            
      Allocatable:             TRUE                               
      Alloc Granule:           4KB                                
      Alloc Alignment:         4KB                                
      Accessible by all:       TRUE                               
  ISA Info:                
*******                  
Agent 3                  
*******                  
  Name:                    gfx90a                             
  Uuid:                    GPU-1901ead3a2ecd4a6               
  Marketing Name:          AMD Instinct MI210                 
  Vendor Name:             AMD                                
  Feature:                 KERNEL_DISPATCH                    
  Profile:                 BASE_PROFILE                       
  Float Round Mode:        NEAR                               
  Max Queue Number:        128(0x80)                          
  Queue Min Size:          64(0x40)                           
  Queue Max Size:          131072(0x20000)                    
  Queue Type:              MULTI                              
  Node:                    2                                  
  Device Type:             GPU                                
  Cache Info:              
    L1:                      16(0x10) KB                        
    L2:                      8192(0x2000) KB                    
  Chip ID:                 29711(0x740f)                      
  ASIC Revision:           1(0x1)                             
  Cacheline Size:          64(0x40)                           
  Max Clock Freq. (MHz):   1700                               
  BDFID:                   25344                              
  Internal Node ID:        2                                  
  Compute Unit:            104                                
  SIMDs per CU:            4                                  
  Shader Engines:          8                                  
  Shader Arrs. per Eng.:   1                                  
  WatchPts on Addr. Ranges:4                                  
  Features:                KERNEL_DISPATCH 
  Fast F16 Operation:      TRUE                               
  Wavefront Size:          64(0x40)                           
  Workgroup Max Size:      1024(0x400)                        
  Workgroup Max Size per Dimension:
    x                        1024(0x400)                        
    y                        1024(0x400)                        
    z                        1024(0x400)                        
  Max Waves Per CU:        32(0x20)                           
  Max Work-item Per CU:    2048(0x800)                        
  Grid Max Size:           4294967295(0xffffffff)             
  Grid Max Size per Dimension:
    x                        4294967295(0xffffffff)             
    y                        4294967295(0xffffffff)             
    z                        4294967295(0xffffffff)             
  Max fbarriers/Workgrp:   32                                 
  Packet Processor uCode:: 78                                 
  SDMA engine uCode::      8                                  
  IOMMU Support::          None                               
  Pool Info:               
    Pool 1                   
      Segment:                 GLOBAL; FLAGS: COARSE GRAINED      
      Size:                    67092480(0x3ffc000) KB             
      Allocatable:             TRUE                               
      Alloc Granule:           4KB                                
      Alloc Alignment:         4KB                                
      Accessible by all:       FALSE                              
    Pool 2                   
      Segment:                 GLOBAL; FLAGS:                     
      Size:                    67092480(0x3ffc000) KB             
      Allocatable:             TRUE                               
      Alloc Granule:           4KB                                
      Alloc Alignment:         4KB                                
      Accessible by all:       FALSE                              
    Pool 3                   
      Segment:                 GLOBAL; FLAGS: FINE GRAINED        
      Size:                    67092480(0x3ffc000) KB             
      Allocatable:             TRUE                               
      Alloc Granule:           4KB                                
      Alloc Alignment:         4KB                                
      Accessible by all:       FALSE                              
    Pool 4                   
      Segment:                 GROUP                              
      Size:                    64(0x40) KB                        
      Allocatable:             FALSE                              
      Alloc Granule:           0KB                                
      Alloc Alignment:         0KB                                
      Accessible by all:       FALSE                              
  ISA Info:                
    ISA 1                    
      Name:                    amdgcn-amd-amdhsa--gfx90a:sramecc+:xnack-
      Machine Models:          HSA_MACHINE_MODEL_LARGE            
      Profiles:                HSA_PROFILE_BASE                   
      Default Rounding Mode:   NEAR                               
      Default Rounding Mode:   NEAR                               
      Fast f16:                TRUE                               
      Workgroup Max Size:      1024(0x400)                        
      Workgroup Max Size per Dimension:
        x                        1024(0x400)                        
        y                        1024(0x400)                        
        z                        1024(0x400)                        
      Grid Max Size:           4294967295(0xffffffff)             
      Grid Max Size per Dimension:
        x                        4294967295(0xffffffff)             
        y                        4294967295(0xffffffff)             
        z                        4294967295(0xffffffff)             
      FBarrier Max Size:       32                                 
*******                  
Agent 4                  
*******                  
  Name:                    gfx90a                             
  Uuid:                    GPU-51276ac12ce05db3               
  Marketing Name:          AMD Instinct MI210                 
  Vendor Name:             AMD                                
  Feature:                 KERNEL_DISPATCH                    
  Profile:                 BASE_PROFILE                       
  Float Round Mode:        NEAR                               
  Max Queue Number:        128(0x80)                          
  Queue Min Size:          64(0x40)                           
  Queue Max Size:          131072(0x20000)                    
  Queue Type:              MULTI                              
  Node:                    3                                  
  Device Type:             GPU                                
  Cache Info:              
    L1:                      16(0x10) KB                        
    L2:                      8192(0x2000) KB                    
  Chip ID:                 29711(0x740f)                      
  ASIC Revision:           1(0x1)                             
  Cacheline Size:          64(0x40)                           
  Max Clock Freq. (MHz):   1700                               
  BDFID:                   17152                              
  Internal Node ID:        3                                  
  Compute Unit:            104                                
  SIMDs per CU:            4                                  
  Shader Engines:          8                                  
  Shader Arrs. per Eng.:   1                                  
  WatchPts on Addr. Ranges:4                                  
  Features:                KERNEL_DISPATCH 
  Fast F16 Operation:      TRUE                               
  Wavefront Size:          64(0x40)                           
  Workgroup Max Size:      1024(0x400)                        
  Workgroup Max Size per Dimension:
    x                        1024(0x400)                        
    y                        1024(0x400)                        
    z                        1024(0x400)                        
  Max Waves Per CU:        32(0x20)                           
  Max Work-item Per CU:    2048(0x800)                        
  Grid Max Size:           4294967295(0xffffffff)             
  Grid Max Size per Dimension:
    x                        4294967295(0xffffffff)             
    y                        4294967295(0xffffffff)             
    z                        4294967295(0xffffffff)             
  Max fbarriers/Workgrp:   32                                 
  Packet Processor uCode:: 78                                 
  SDMA engine uCode::      8                                  
  IOMMU Support::          None                               
  Pool Info:               
    Pool 1                   
      Segment:                 GLOBAL; FLAGS: COARSE GRAINED      
      Size:                    67092480(0x3ffc000) KB             
      Allocatable:             TRUE                               
      Alloc Granule:           4KB                                
      Alloc Alignment:         4KB                                
      Accessible by all:       FALSE                              
    Pool 2                   
      Segment:                 GLOBAL; FLAGS:                     
      Size:                    67092480(0x3ffc000) KB             
      Allocatable:             TRUE                               
      Alloc Granule:           4KB                                
      Alloc Alignment:         4KB                                
      Accessible by all:       FALSE                              
    Pool 3                   
      Segment:                 GLOBAL; FLAGS: FINE GRAINED        
      Size:                    67092480(0x3ffc000) KB             
      Allocatable:             TRUE                               
      Alloc Granule:           4KB                                
      Alloc Alignment:         4KB                                
      Accessible by all:       FALSE                              
    Pool 4                   
      Segment:                 GROUP                              
      Size:                    64(0x40) KB                        
      Allocatable:             FALSE                              
      Alloc Granule:           0KB                                
      Alloc Alignment:         0KB                                
      Accessible by all:       FALSE                              
  ISA Info:                
    ISA 1                    
      Name:                    amdgcn-amd-amdhsa--gfx90a:sramecc+:xnack-
      Machine Models:          HSA_MACHINE_MODEL_LARGE            
      Profiles:                HSA_PROFILE_BASE                   
      Default Rounding Mode:   NEAR                               
      Default Rounding Mode:   NEAR                               
      Fast f16:                TRUE                               
      Workgroup Max Size:      1024(0x400)                        
      Workgroup Max Size per Dimension:
        x                        1024(0x400)                        
        y                        1024(0x400)                        
        z                        1024(0x400)                        
      Grid Max Size:           4294967295(0xffffffff)             
      Grid Max Size per Dimension:
        x                        4294967295(0xffffffff)             
        y                        4294967295(0xffffffff)             
        z                        4294967295(0xffffffff)             
      FBarrier Max Size:       32                                 
*******                  
Agent 5                  
*******                  
  Name:                    gfx90a                             
  Uuid:                    GPU-93bb3629c01aea1c               
  Marketing Name:          AMD Instinct MI210                 
  Vendor Name:             AMD                                
  Feature:                 KERNEL_DISPATCH                    
  Profile:                 BASE_PROFILE                       
  Float Round Mode:        NEAR                               
  Max Queue Number:        128(0x80)                          
  Queue Min Size:          64(0x40)                           
  Queue Max Size:          131072(0x20000)                    
  Queue Type:              MULTI                              
  Node:                    4                                  
  Device Type:             GPU                                
  Cache Info:              
    L1:                      16(0x10) KB                        
    L2:                      8192(0x2000) KB                    
  Chip ID:                 29711(0x740f)                      
  ASIC Revision:           1(0x1)                             
  Cacheline Size:          64(0x40)                           
  Max Clock Freq. (MHz):   1700                               
  BDFID:                   768                                
  Internal Node ID:        4                                  
  Compute Unit:            104                                
  SIMDs per CU:            4                                  
  Shader Engines:          8                                  
  Shader Arrs. per Eng.:   1                                  
  WatchPts on Addr. Ranges:4                                  
  Features:                KERNEL_DISPATCH 
  Fast F16 Operation:      TRUE                               
  Wavefront Size:          64(0x40)                           
  Workgroup Max Size:      1024(0x400)                        
  Workgroup Max Size per Dimension:
    x                        1024(0x400)                        
    y                        1024(0x400)                        
    z                        1024(0x400)                        
  Max Waves Per CU:        32(0x20)                           
  Max Work-item Per CU:    2048(0x800)                        
  Grid Max Size:           4294967295(0xffffffff)             
  Grid Max Size per Dimension:
    x                        4294967295(0xffffffff)             
    y                        4294967295(0xffffffff)             
    z                        4294967295(0xffffffff)             
  Max fbarriers/Workgrp:   32                                 
  Packet Processor uCode:: 78                                 
  SDMA engine uCode::      8                                  
  IOMMU Support::          None                               
  Pool Info:               
    Pool 1                   
      Segment:                 GLOBAL; FLAGS: COARSE GRAINED      
      Size:                    67092480(0x3ffc000) KB             
      Allocatable:             TRUE                               
      Alloc Granule:           4KB                                
      Alloc Alignment:         4KB                                
      Accessible by all:       FALSE                              
    Pool 2                   
      Segment:                 GLOBAL; FLAGS:                     
      Size:                    67092480(0x3ffc000) KB             
      Allocatable:             TRUE                               
      Alloc Granule:           4KB                                
      Alloc Alignment:         4KB                                
      Accessible by all:       FALSE                              
    Pool 3                   
      Segment:                 GLOBAL; FLAGS: FINE GRAINED        
      Size:                    67092480(0x3ffc000) KB             
      Allocatable:             TRUE                               
      Alloc Granule:           4KB                                
      Alloc Alignment:         4KB                                
      Accessible by all:       FALSE                              
    Pool 4                   
      Segment:                 GROUP                              
      Size:                    64(0x40) KB                        
      Allocatable:             FALSE                              
      Alloc Granule:           0KB                                
      Alloc Alignment:         0KB                                
      Accessible by all:       FALSE                              
  ISA Info:                
    ISA 1                    
      Name:                    amdgcn-amd-amdhsa--gfx90a:sramecc+:xnack-
      Machine Models:          HSA_MACHINE_MODEL_LARGE            
      Profiles:                HSA_PROFILE_BASE                   
      Default Rounding Mode:   NEAR                               
      Default Rounding Mode:   NEAR                               
      Fast f16:                TRUE                               
      Workgroup Max Size:      1024(0x400)                        
      Workgroup Max Size per Dimension:
        x                        1024(0x400)                        
        y                        1024(0x400)                        
        z                        1024(0x400)                        
      Grid Max Size:           4294967295(0xffffffff)             
      Grid Max Size per Dimension:
        x                        4294967295(0xffffffff)             
        y                        4294967295(0xffffffff)             
        z                        4294967295(0xffffffff)             
      FBarrier Max Size:       32                                 
*******                  
Agent 6                  
*******                  
  Name:                    gfx90a                             
  Uuid:                    GPU-4344e78b718814cd               
  Marketing Name:          AMD Instinct MI210                 
  Vendor Name:             AMD                                
  Feature:                 KERNEL_DISPATCH                    
  Profile:                 BASE_PROFILE                       
  Float Round Mode:        NEAR                               
  Max Queue Number:        128(0x80)                          
  Queue Min Size:          64(0x40)                           
  Queue Max Size:          131072(0x20000)                    
  Queue Type:              MULTI                              
  Node:                    5                                  
  Device Type:             GPU                                
  Cache Info:              
    L1:                      16(0x10) KB                        
    L2:                      8192(0x2000) KB                    
  Chip ID:                 29711(0x740f)                      
  ASIC Revision:           1(0x1)                             
  Cacheline Size:          64(0x40)                           
  Max Clock Freq. (MHz):   1700                               
  BDFID:                   9728                               
  Internal Node ID:        5                                  
  Compute Unit:            104                                
  SIMDs per CU:            4                                  
  Shader Engines:          8                                  
  Shader Arrs. per Eng.:   1                                  
  WatchPts on Addr. Ranges:4                                  
  Features:                KERNEL_DISPATCH 
  Fast F16 Operation:      TRUE                               
  Wavefront Size:          64(0x40)                           
  Workgroup Max Size:      1024(0x400)                        
  Workgroup Max Size per Dimension:
    x                        1024(0x400)                        
    y                        1024(0x400)                        
    z                        1024(0x400)                        
  Max Waves Per CU:        32(0x20)                           
  Max Work-item Per CU:    2048(0x800)                        
  Grid Max Size:           4294967295(0xffffffff)             
  Grid Max Size per Dimension:
    x                        4294967295(0xffffffff)             
    y                        4294967295(0xffffffff)             
    z                        4294967295(0xffffffff)             
  Max fbarriers/Workgrp:   32                                 
  Packet Processor uCode:: 78                                 
  SDMA engine uCode::      8                                  
  IOMMU Support::          None                               
  Pool Info:               
    Pool 1                   
      Segment:                 GLOBAL; FLAGS: COARSE GRAINED      
      Size:                    67092480(0x3ffc000) KB             
      Allocatable:             TRUE                               
      Alloc Granule:           4KB                                
      Alloc Alignment:         4KB                                
      Accessible by all:       FALSE                              
    Pool 2                   
      Segment:                 GLOBAL; FLAGS:                     
      Size:                    67092480(0x3ffc000) KB             
      Allocatable:             TRUE                               
      Alloc Granule:           4KB                                
      Alloc Alignment:         4KB                                
      Accessible by all:       FALSE                              
    Pool 3                   
      Segment:                 GLOBAL; FLAGS: FINE GRAINED        
      Size:                    67092480(0x3ffc000) KB             
      Allocatable:             TRUE                               
      Alloc Granule:           4KB                                
      Alloc Alignment:         4KB                                
      Accessible by all:       FALSE                              
    Pool 4                   
      Segment:                 GROUP                              
      Size:                    64(0x40) KB                        
      Allocatable:             FALSE                              
      Alloc Granule:           0KB                                
      Alloc Alignment:         0KB                                
      Accessible by all:       FALSE                              
  ISA Info:                
    ISA 1                    
      Name:                    amdgcn-amd-amdhsa--gfx90a:sramecc+:xnack-
      Machine Models:          HSA_MACHINE_MODEL_LARGE            
      Profiles:                HSA_PROFILE_BASE                   
      Default Rounding Mode:   NEAR                               
      Default Rounding Mode:   NEAR                               
      Fast f16:                TRUE                               
      Workgroup Max Size:      1024(0x400)                        
      Workgroup Max Size per Dimension:
        x                        1024(0x400)                        
        y                        1024(0x400)                        
        z                        1024(0x400)                        
      Grid Max Size:           4294967295(0xffffffff)             
      Grid Max Size per Dimension:
        x                        4294967295(0xffffffff)             
        y                        4294967295(0xffffffff)             
        z                        4294967295(0xffffffff)             
      FBarrier Max Size:       32                                 
*******                  
Agent 7                  
*******                  
  Name:                    gfx90a                             
  Uuid:                    GPU-aea834aa2723eb57               
  Marketing Name:          AMD Instinct MI210                 
  Vendor Name:             AMD                                
  Feature:                 KERNEL_DISPATCH                    
  Profile:                 BASE_PROFILE                       
  Float Round Mode:        NEAR                               
  Max Queue Number:        128(0x80)                          
  Queue Min Size:          64(0x40)                           
  Queue Max Size:          131072(0x20000)                    
  Queue Type:              MULTI                              
  Node:                    6                                  
  Device Type:             GPU                                
  Cache Info:              
    L1:                      16(0x10) KB                        
    L2:                      8192(0x2000) KB                    
  Chip ID:                 29711(0x740f)                      
  ASIC Revision:           1(0x1)                             
  Cacheline Size:          64(0x40)                           
  Max Clock Freq. (MHz):   1700                               
  BDFID:                   58112                              
  Internal Node ID:        6                                  
  Compute Unit:            104                                
  SIMDs per CU:            4                                  
  Shader Engines:          8                                  
  Shader Arrs. per Eng.:   1                                  
  WatchPts on Addr. Ranges:4                                  
  Features:                KERNEL_DISPATCH 
  Fast F16 Operation:      TRUE                               
  Wavefront Size:          64(0x40)                           
  Workgroup Max Size:      1024(0x400)                        
  Workgroup Max Size per Dimension:
    x                        1024(0x400)                        
    y                        1024(0x400)                        
    z                        1024(0x400)                        
  Max Waves Per CU:        32(0x20)                           
  Max Work-item Per CU:    2048(0x800)                        
  Grid Max Size:           4294967295(0xffffffff)             
  Grid Max Size per Dimension:
    x                        4294967295(0xffffffff)             
    y                        4294967295(0xffffffff)             
    z                        4294967295(0xffffffff)             
  Max fbarriers/Workgrp:   32                                 
  Packet Processor uCode:: 78                                 
  SDMA engine uCode::      8                                  
  IOMMU Support::          None                               
  Pool Info:               
    Pool 1                   
      Segment:                 GLOBAL; FLAGS: COARSE GRAINED      
      Size:                    67092480(0x3ffc000) KB             
      Allocatable:             TRUE                               
      Alloc Granule:           4KB                                
      Alloc Alignment:         4KB                                
      Accessible by all:       FALSE                              
    Pool 2                   
      Segment:                 GLOBAL; FLAGS:                     
      Size:                    67092480(0x3ffc000) KB             
      Allocatable:             TRUE                               
      Alloc Granule:           4KB                                
      Alloc Alignment:         4KB                                
      Accessible by all:       FALSE                              
    Pool 3                   
      Segment:                 GLOBAL; FLAGS: FINE GRAINED        
      Size:                    67092480(0x3ffc000) KB             
      Allocatable:             TRUE                               
      Alloc Granule:           4KB                                
      Alloc Alignment:         4KB                                
      Accessible by all:       FALSE                              
    Pool 4                   
      Segment:                 GROUP                              
      Size:                    64(0x40) KB                        
      Allocatable:             FALSE                              
      Alloc Granule:           0KB                                
      Alloc Alignment:         0KB                                
      Accessible by all:       FALSE                              
  ISA Info:                
    ISA 1                    
      Name:                    amdgcn-amd-amdhsa--gfx90a:sramecc+:xnack-
      Machine Models:          HSA_MACHINE_MODEL_LARGE            
      Profiles:                HSA_PROFILE_BASE                   
      Default Rounding Mode:   NEAR                               
      Default Rounding Mode:   NEAR                               
      Fast f16:                TRUE                               
      Workgroup Max Size:      1024(0x400)                        
      Workgroup Max Size per Dimension:
        x                        1024(0x400)                        
        y                        1024(0x400)                        
        z                        1024(0x400)                        
      Grid Max Size:           4294967295(0xffffffff)             
      Grid Max Size per Dimension:
        x                        4294967295(0xffffffff)             
        y                        4294967295(0xffffffff)             
        z                        4294967295(0xffffffff)             
      FBarrier Max Size:       32                                 
*******                  
Agent 8                  
*******                  
  Name:                    gfx90a                             
  Uuid:                    GPU-a3781b5a10fcf54a               
  Marketing Name:          AMD Instinct MI210                 
  Vendor Name:             AMD                                
  Feature:                 KERNEL_DISPATCH                    
  Profile:                 BASE_PROFILE                       
  Float Round Mode:        NEAR                               
  Max Queue Number:        128(0x80)                          
  Queue Min Size:          64(0x40)                           
  Queue Max Size:          131072(0x20000)                    
  Queue Type:              MULTI                              
  Node:                    7                                  
  Device Type:             GPU                                
  Cache Info:              
    L1:                      16(0x10) KB                        
    L2:                      8192(0x2000) KB                    
  Chip ID:                 29711(0x740f)                      
  ASIC Revision:           1(0x1)                             
  Cacheline Size:          64(0x40)                           
  Max Clock Freq. (MHz):   1700                               
  BDFID:                   49920                              
  Internal Node ID:        7                                  
  Compute Unit:            104                                
  SIMDs per CU:            4                                  
  Shader Engines:          8                                  
  Shader Arrs. per Eng.:   1                                  
  WatchPts on Addr. Ranges:4                                  
  Features:                KERNEL_DISPATCH 
  Fast F16 Operation:      TRUE                               
  Wavefront Size:          64(0x40)                           
  Workgroup Max Size:      1024(0x400)                        
  Workgroup Max Size per Dimension:
    x                        1024(0x400)                        
    y                        1024(0x400)                        
    z                        1024(0x400)                        
  Max Waves Per CU:        32(0x20)                           
  Max Work-item Per CU:    2048(0x800)                        
  Grid Max Size:           4294967295(0xffffffff)             
  Grid Max Size per Dimension:
    x                        4294967295(0xffffffff)             
    y                        4294967295(0xffffffff)             
    z                        4294967295(0xffffffff)             
  Max fbarriers/Workgrp:   32                                 
  Packet Processor uCode:: 78                                 
  SDMA engine uCode::      8                                  
  IOMMU Support::          None                               
  Pool Info:               
    Pool 1                   
      Segment:                 GLOBAL; FLAGS: COARSE GRAINED      
      Size:                    67092480(0x3ffc000) KB             
      Allocatable:             TRUE                               
      Alloc Granule:           4KB                                
      Alloc Alignment:         4KB                                
      Accessible by all:       FALSE                              
    Pool 2                   
      Segment:                 GLOBAL; FLAGS:                     
      Size:                    67092480(0x3ffc000) KB             
      Allocatable:             TRUE                               
      Alloc Granule:           4KB                                
      Alloc Alignment:         4KB                                
      Accessible by all:       FALSE                              
    Pool 3                   
      Segment:                 GLOBAL; FLAGS: FINE GRAINED        
      Size:                    67092480(0x3ffc000) KB             
      Allocatable:             TRUE                               
      Alloc Granule:           4KB                                
      Alloc Alignment:         4KB                                
      Accessible by all:       FALSE                              
    Pool 4                   
      Segment:                 GROUP                              
      Size:                    64(0x40) KB                        
      Allocatable:             FALSE                              
      Alloc Granule:           0KB                                
      Alloc Alignment:         0KB                                
      Accessible by all:       FALSE                              
  ISA Info:                
    ISA 1                    
      Name:                    amdgcn-amd-amdhsa--gfx90a:sramecc+:xnack-
      Machine Models:          HSA_MACHINE_MODEL_LARGE            
      Profiles:                HSA_PROFILE_BASE                   
      Default Rounding Mode:   NEAR                               
      Default Rounding Mode:   NEAR                               
      Fast f16:                TRUE                               
      Workgroup Max Size:      1024(0x400)                        
      Workgroup Max Size per Dimension:
        x                        1024(0x400)                        
        y                        1024(0x400)                        
        z                        1024(0x400)                        
      Grid Max Size:           4294967295(0xffffffff)             
      Grid Max Size per Dimension:
        x                        4294967295(0xffffffff)             
        y                        4294967295(0xffffffff)             
        z                        4294967295(0xffffffff)             
      FBarrier Max Size:       32                                 
*******                  
Agent 9                  
*******                  
  Name:                    gfx90a                             
  Uuid:                    GPU-f83f06efb714e6cf               
  Marketing Name:          AMD Instinct MI210                 
  Vendor Name:             AMD                                
  Feature:                 KERNEL_DISPATCH                    
  Profile:                 BASE_PROFILE                       
  Float Round Mode:        NEAR                               
  Max Queue Number:        128(0x80)                          
  Queue Min Size:          64(0x40)                           
  Queue Max Size:          131072(0x20000)                    
  Queue Type:              MULTI                              
  Node:                    8                                  
  Device Type:             GPU                                
  Cache Info:              
    L1:                      16(0x10) KB                        
    L2:                      8192(0x2000) KB                    
  Chip ID:                 29711(0x740f)                      
  ASIC Revision:           1(0x1)                             
  Cacheline Size:          64(0x40)                           
  Max Clock Freq. (MHz):   1700                               
  BDFID:                   33536                              
  Internal Node ID:        8                                  
  Compute Unit:            104                                
  SIMDs per CU:            4                                  
  Shader Engines:          8                                  
  Shader Arrs. per Eng.:   1                                  
  WatchPts on Addr. Ranges:4                                  
  Features:                KERNEL_DISPATCH 
  Fast F16 Operation:      TRUE                               
  Wavefront Size:          64(0x40)                           
  Workgroup Max Size:      1024(0x400)                        
  Workgroup Max Size per Dimension:
    x                        1024(0x400)                        
    y                        1024(0x400)                        
    z                        1024(0x400)                        
  Max Waves Per CU:        32(0x20)                           
  Max Work-item Per CU:    2048(0x800)                        
  Grid Max Size:           4294967295(0xffffffff)             
  Grid Max Size per Dimension:
    x                        4294967295(0xffffffff)             
    y                        4294967295(0xffffffff)             
    z                        4294967295(0xffffffff)             
  Max fbarriers/Workgrp:   32                                 
  Packet Processor uCode:: 78                                 
  SDMA engine uCode::      8                                  
  IOMMU Support::          None                               
  Pool Info:               
    Pool 1                   
      Segment:                 GLOBAL; FLAGS: COARSE GRAINED      
      Size:                    67092480(0x3ffc000) KB             
      Allocatable:             TRUE                               
      Alloc Granule:           4KB                                
      Alloc Alignment:         4KB                                
      Accessible by all:       FALSE                              
    Pool 2                   
      Segment:                 GLOBAL; FLAGS:                     
      Size:                    67092480(0x3ffc000) KB             
      Allocatable:             TRUE                               
      Alloc Granule:           4KB                                
      Alloc Alignment:         4KB                                
      Accessible by all:       FALSE                              
    Pool 3                   
      Segment:                 GLOBAL; FLAGS: FINE GRAINED        
      Size:                    67092480(0x3ffc000) KB             
      Allocatable:             TRUE                               
      Alloc Granule:           4KB                                
      Alloc Alignment:         4KB                                
      Accessible by all:       FALSE                              
    Pool 4                   
      Segment:                 GROUP                              
      Size:                    64(0x40) KB                        
      Allocatable:             FALSE                              
      Alloc Granule:           0KB                                
      Alloc Alignment:         0KB                                
      Accessible by all:       FALSE                              
  ISA Info:                
    ISA 1                    
      Name:                    amdgcn-amd-amdhsa--gfx90a:sramecc+:xnack-
      Machine Models:          HSA_MACHINE_MODEL_LARGE            
      Profiles:                HSA_PROFILE_BASE                   
      Default Rounding Mode:   NEAR                               
      Default Rounding Mode:   NEAR                               
      Fast f16:                TRUE                               
      Workgroup Max Size:      1024(0x400)                        
      Workgroup Max Size per Dimension:
        x                        1024(0x400)                        
        y                        1024(0x400)                        
        z                        1024(0x400)                        
      Grid Max Size:           4294967295(0xffffffff)             
      Grid Max Size per Dimension:
        x                        4294967295(0xffffffff)             
        y                        4294967295(0xffffffff)             
        z                        4294967295(0xffffffff)             
      FBarrier Max Size:       32                                 
*******                  
Agent 10                 
*******                  
  Name:                    gfx90a                             
  Uuid:                    GPU-c34ec50444dd0a6d               
  Marketing Name:          AMD Instinct MI210                 
  Vendor Name:             AMD                                
  Feature:                 KERNEL_DISPATCH                    
  Profile:                 BASE_PROFILE                       
  Float Round Mode:        NEAR                               
  Max Queue Number:        128(0x80)                          
  Queue Min Size:          64(0x40)                           
  Queue Max Size:          131072(0x20000)                    
  Queue Type:              MULTI                              
  Node:                    9                                  
  Device Type:             GPU                                
  Cache Info:              
    L1:                      16(0x10) KB                        
    L2:                      8192(0x2000) KB                    
  Chip ID:                 29711(0x740f)                      
  ASIC Revision:           1(0x1)                             
  Cacheline Size:          64(0x40)                           
  Max Clock Freq. (MHz):   1700                               
  BDFID:                   41728                              
  Internal Node ID:        9                                  
  Compute Unit:            104                                
  SIMDs per CU:            4                                  
  Shader Engines:          8                                  
  Shader Arrs. per Eng.:   1                                  
  WatchPts on Addr. Ranges:4                                  
  Features:                KERNEL_DISPATCH 
  Fast F16 Operation:      TRUE                               
  Wavefront Size:          64(0x40)                           
  Workgroup Max Size:      1024(0x400)                        
  Workgroup Max Size per Dimension:
    x                        1024(0x400)                        
    y                        1024(0x400)                        
    z                        1024(0x400)                        
  Max Waves Per CU:        32(0x20)                           
  Max Work-item Per CU:    2048(0x800)                        
  Grid Max Size:           4294967295(0xffffffff)             
  Grid Max Size per Dimension:
    x                        4294967295(0xffffffff)             
    y                        4294967295(0xffffffff)             
    z                        4294967295(0xffffffff)             
  Max fbarriers/Workgrp:   32                                 
  Packet Processor uCode:: 78                                 
  SDMA engine uCode::      8                                  
  IOMMU Support::          None                               
  Pool Info:               
    Pool 1                   
      Segment:                 GLOBAL; FLAGS: COARSE GRAINED      
      Size:                    67092480(0x3ffc000) KB             
      Allocatable:             TRUE                               
      Alloc Granule:           4KB                                
      Alloc Alignment:         4KB                                
      Accessible by all:       FALSE                              
    Pool 2                   
      Segment:                 GLOBAL; FLAGS:                     
      Size:                    67092480(0x3ffc000) KB             
      Allocatable:             TRUE                               
      Alloc Granule:           4KB                                
      Alloc Alignment:         4KB                                
      Accessible by all:       FALSE                              
    Pool 3                   
      Segment:                 GLOBAL; FLAGS: FINE GRAINED        
      Size:                    67092480(0x3ffc000) KB             
      Allocatable:             TRUE                               
      Alloc Granule:           4KB                                
      Alloc Alignment:         4KB                                
      Accessible by all:       FALSE                              
    Pool 4                   
      Segment:                 GROUP                              
      Size:                    64(0x40) KB                        
      Allocatable:             FALSE                              
      Alloc Granule:           0KB                                
      Alloc Alignment:         0KB                                
      Accessible by all:       FALSE                              
  ISA Info:                
    ISA 1                    
      Name:                    amdgcn-amd-amdhsa--gfx90a:sramecc+:xnack-
      Machine Models:          HSA_MACHINE_MODEL_LARGE            
      Profiles:                HSA_PROFILE_BASE                   
      Default Rounding Mode:   NEAR                               
      Default Rounding Mode:   NEAR                               
      Fast f16:                TRUE                               
      Workgroup Max Size:      1024(0x400)                        
      Workgroup Max Size per Dimension:
        x                        1024(0x400)                        
        y                        1024(0x400)                        
        z                        1024(0x400)                        
      Grid Max Size:           4294967295(0xffffffff)             
      Grid Max Size per Dimension:
        x                        4294967295(0xffffffff)             
        y                        4294967295(0xffffffff)             
        z                        4294967295(0xffffffff)             
      FBarrier Max Size:       32                                 
*** Done ***    

### Additional Information

_No response_
@donglixp
Copy link

donglixp commented Feb 3, 2024

The similar error: #41

@dejay-vu
Copy link

dejay-vu commented Feb 7, 2024

@ehartford, can I have the scripts you are running?

@ehartford
Copy link
Author

I will try to define the configuration so you can reproduce it

@seungduk-yanolja
Copy link

+1

https://github.com/ROCm/flash-attention/blob/flash_attention_for_rocm/csrc/flash_attn_rocm/src/params.hpp#L450
here, it checks if they are the same but my log printed

dout_seq_stride: 4096, out_seq_stride: 128

and I modified this code to print some logs
https://github.com/ROCm/flash-attention/blob/flash_attention_for_rocm/flash_attn/flash_attn_interface.py#L73
like this

def _flash_attn_varlen_backward(dout, q, k, v, out, softmax_lse, dq, dk, dv,
                                cu_seqlens_q, cu_seqlens_k, max_seqlen_q, max_seqlen_k,
                                dropout_p, softmax_scale, causal, rng_state=None):

    print("Input tensor (dout) shape:", dout.shape)
    print("Input tensor (dout) strides:", dout.stride())
    print("Output tensor (out) shape:", out.shape)
    print("Output tensor (out) strides:", out.stride())

    maybe_contiguous = lambda x: x.contiguous() if x.stride(-1) != 1 else x
    # dq, dk, dv are allocated by us so they should already be contiguous
    dout, q, k, v, out = [maybe_contiguous(x) for x in (dout, q, k, v, out)]


    print("After Input tensor (dout) shape:", dout.shape)
    print("After Input tensor (dout) strides:", dout.stride())
    print("After Output tensor (out) shape:", out.shape)
    print("After Output tensor (out) strides:", out.stride())

    dq, dk, dv, softmax_d, = flash_attn_cuda.varlen_bwd(
        dout, q, k, v, out, softmax_lse, dq, dk, dv, cu_seqlens_q, cu_seqlens_k,
        max_seqlen_q, max_seqlen_k, dropout_p, softmax_scale, False, causal, None, rng_state
    )
    # if dk.isnan().any() or dk.isnan().any() or dv.isnan().any() or softmax_d.isnan().any():
    #     breakpoint()
    return dq, dk, dv, softmax_d

and result

Input tensor (dout) shape: torch.Size([32768, 32, 128])
Input tensor (dout) strides: (4096, 128, 1)
Output tensor (out) shape: torch.Size([32768, 32, 128])
Output tensor (out) strides: (128, 4194304, 1)
After Input tensor (dout) shape: torch.Size([32768, 32, 128])
After Input tensor (dout) strides: (4096, 128, 1)
After Output tensor (out) shape: torch.Size([32768, 32, 128])
After Output tensor (out) strides: (128, 4194304, 1)

@seungduk-yanolja seungduk-yanolja mentioned this issue Apr 5, 2024
Closed
@jamestwhedbee
Copy link

Running into this as well

@ehartford
Copy link
Author

I just discovered it again

RuntimeError: Expected dout_seq_stride == out_seq_stride to be true, but got false.  (Could this error message be improved?  If so, please report an enhancement request to PyTorch.)
  File "/home/ehartford/miniconda3/envs/axolotl/lib/python3.11/site-packages/torch/autograd/__init__.py", line 251, in backward
    _flash_attn_varlen_backward(
  File "/home/ehartford/miniconda3/envs/axolotl/lib/python3.11/site-packages/flash_attn/flash_attn_interface.py", line 79, in _flash_attn_varlen_backward
    dq, dk, dv, softmax_d, = flash_attn_cuda.varlen_bwd(
    Variable._execution_engine.run_backward(  # Calls into the C++ engine to run the backward pass 
    File "/home/ehartford/miniconda3/envs/axolotl/lib/python3.11/site-packages/torch/autograd/function.py", line 288, in apply
                            return user_fn(self, *args) 
 ^^^^^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^^^^^^^^^^^^^^^^^^^^^^^^
^^^^RuntimeError^^: ^Expected dout_seq_stride == out_seq_stride to be true, but got false.  (Could this error message be improved?  If so, please report an enhancement request to PyTorch.)^

  File "/home/ehartford/miniconda3/envs/axolotl/lib/python3.11/site-packages/flash_attn/flash_attn_interface.py", line 139, in backward
    _flash_attn_varlen_backward(
  File "/home/ehartford/miniconda3/envs/axolotl/lib/python3.11/site-packages/flash_attn/flash_attn_interface.py", line 79, in _flash_attn_varlen_backward
    dq, dk, dv, softmax_d, = flash_attn_cuda.varlen_bwd(
                             ^^^^^^^^^^^^^^^^^^^^^^^^^^^

@ehartford
Copy link
Author

ehartford commented May 27, 2024 •
edited
Loading

Hello, I am blocked from using flash attention on ROCm by this bug

@ehartford ehartford mentioned this issue May 27, 2024
Closed
@ehartford
Copy link
Author

This still happening with ROCm 6.0.2

@ehartford
Copy link
Author

gfx942 mi300x

@NathanGodey
Copy link

I just ran into the same issue (MI250x) when using GPTNeoXForCausalLM from HuggingFace. I could fix it by just making all the inputs contiguous right before the flash-attention call, which I guess is suboptimal but still much better than the torch implementation.

@clintg6 clintg6 linked a pull request May 31, 2024 that will close this issue
Open
@ehartford
Copy link
Author

I will try

@rocking5566
Copy link
Collaborator

I just submit an PR to support AMD / ROCm on FlashAttention 2
Dao-AILab#1010
This PR using composable_kernel as backend

@ehartford
Copy link
Author

I will try this

@carlushuang
Copy link
Collaborator

cc @danyao12

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

Successfully merging a pull request may close this issue.

Fix stride issues in flash_attn_interface ROCm/flash-attention
8 participants
@donglixp @ehartford @carlushuang @rocking5566 @jamestwhedbee @NathanGodey @dejay-vu @seungduk-yanolja