Skip to content

Multilayer models #6

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
Draft
alexfrater wants to merge 72 commits into
base: main
Choose a base branch
from
Draft

Multilayer models #6

alexfrater wants to merge 72 commits into from

Conversation

@alexfrater
Copy link
Contributor

@alexfrater alexfrater commented Jul 27, 2024 •
edited
Loading

Enabled support for multi-layered models

  • Added mapping for in and out messages between different layers using HBM as temporary storage
  • Added layer config parameter to enable aggregation without reprogramming
  • Added MLP Model
  • Added GCN model with linear layer (used in Graphcast) that can be run on AMPLE without reprogramming nodeslots between GCN and linear layers
  • Added software to change adjancey list between layers

Enabled support for edge embedding

Enabled support for Interaction Net

https://arxiv.org/pdf/1612.00222

Replaced Xilinx IP for simulation

  • Work in progress - Need to rewrite using UG902

Verilator Makefile

  • Work in progress - Not compiling correctly due to warnings

Multilayer Transform Instruction

image

Concatenation Transform Instruction

image

image

###Memory Map

image

GCN with Concatenation Instruction

image

alexfrater added 30 commits June 12, 2024 13:48
@alexfrater
build bug fix
68089ce
@alexfrater
SIM Wave Windows
ba68158
@alexfrater
Temp Sys Path
dd7c3ca
@alexfrater
Fixed Weight Bank Bug
75a700d
@alexfrater
Simualtion Files and MLP model
63dcaca
@alexfrater
Debugging Full Simulation
40dcb98
@alexfrater
functions to save JIT model and graph
fd90818
@alexfrater
gnn model software integration in testbench
93b1b22
@alexfrater
9 jul
d1729b1
@alexfrater
nodeslot writeback monitor checker
a803129
@alexfrater
bug fix: empty weight FIFOs set to full and dumping undefined values …
dfabf34 
…when reset
@alexfrater
gating non-valid fp input values to prevent x-propogation
d50adb4
@alexfrater
fte writeback logic
239440c
@alexfrater
sim files
7539061
@alexfrater
bug generating 4 meshes
91c17d0
@alexfrater
bug generating 4 meshes 2
4e9905e
@alexfrater
aggregation feature write count - enable feature counts of arbitrary …
5b8f72b 
…length
@alexfrater
correct feature input order
31b372d
@alexfrater
testbench output data and size comparision
f3217d1
@alexfrater
fix wrong neigbour message address
38d071c
@alexfrater
allow writeback of non fully populated fte
f8227a7
@alexfrater
fast multiplication bug fix
aaa4300
@alexfrater
valid row fix for nodeslot count greater than tfe channel count (test…
5577d9a 
…bench passed)
@alexfrater
simulation edits
e2ebdac
@alexfrater
bug fix buffer slot reset after transform
7b80cb2
@alexfrater
fix bug last row of fte not reset
5b99d67
@alexfrater
backpressure enabled on agc feature aggregator and bug fix valid sign…
36fadb4 
…al feature aggregator
@alexfrater
mapping memory for varying feature widths in multi-layer mlp
f4c766c
@alexfrater
FTE writeback to HBM
a65f167
@alexfrater
multi layer weights mapping in sdk
6a3c4d9
alexfrater and others added 30 commits July 21, 2024 23:37
@alexfrater
lint3
356a30e
@alexfrater
Update verilog_linting.yml
10061d7
@alexfrater
fix cocotb modelsim error
06f3833
@alexfrater
updated tb - fix bias removal for arbitrary layers
04d60e2
@alexfrater
fixed gcn aggregation bug
735bf43
@alexfrater
added layer config to enable aggregation (issues with some features i…
ba288a2 
…n multi layer GCNs)
@alexfrater
fpga clock cycle benchmarking
be51854
@alexfrater
dynamically add workarea path
ef92b44
@alexfrater
requirments system agnostic
40d7a42
@alexfrater
mkl service
2ec25f0
@alexfrater
added model tb precision
34e2c29
@alexfrater
added arguments to tb and changed logging levels
e0a3c69
@alexfrater
minor changes
c45750d
@alexfrater
cleaning up code 1
22233e6
@alexfrater
cleaning up code 2
333582d
@alexfrater
Merge pull request #3 from alexfrater/mlp-feature-branch
5e1cc86 
Clean Code
@alexfrater
updated gitignore and READM
68ad548
@alexfrater
remove imports
72dbffb
@alexfrater
working adj list offset for each layer
660396c
@alexfrater
removed redunant signals
0fb028f
@alexfrater
fix_bug
6a38e41
@alexfrater
added precision to gcn_mlp
e91769e
@alexfrater
updated sim
cd8418f
@alexfrater
debugging
04bef35
more trials in gpu bmark
6e522e3
@alexfrater
brand publish
be024bb
@alexfrater
temp fixes from float_silu branch to allow verilator compilation
6800fc1
@alexfrater
Merge branch 'multilayer_models' of https://github.com/alexfrater/agile…
d66de93 
… into multilayer_models
@alexfrater
makefile
5d74b9a
@alexfrater
Merge branch 'edge_embeddings' into multilayer_models
a4c2de6 
Added edge embedding models
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

1 participant

@alexfrater