Commits

Breadcrumbs

History for

onmain

User selector

All users

Datepicker

All time

Commit History

Commits on Feb 5, 2025

• delete

  liyin2015

  committed
  9c5c4a7

  Copy full SHA for 9c5c4a7

• improve the clarity and minor bug fix

  liyin2015

  committed
  99c551b

  Copy full SHA for 99c551b

• clean up printc in the training process

  liyin2015

  committed
  935c273

  Copy full SHA for 935c273

Commits on Feb 1, 2025

• change lightrag to adalflow, delete many outdated files

  liyin2015

  committed
  03df847

  Copy full SHA for 03df847

Commits on Jan 29, 2025

• review all changes and improve the code

  liyin2015

  committed
  96179ed

  Copy full SHA for 96179ed

Commits on Jan 28, 2025

• finalize the code

  liyin2015

  committed
  231cca5

  Copy full SHA for 231cca5

Commits on Jan 27, 2025

• last commit

  liyin2015

  committed
  c058d74

  Copy full SHA for c058d74

Commits on Jan 25, 2025

• fix all tests

  liyin2015

  committed
  2d50c43

  Copy full SHA for 2d50c43

Commits on Jan 24, 2025

• enabling programs to disable gradients or the whole backward for cheaper training

  liyin2015

  committed
  b86b543

  Copy full SHA for b86b543

Commits on Jan 22, 2025

• consolidate the gradcomponent and gradcomponent2 into one

  liyin2015

  committed
  fff00d5

  Copy full SHA for fff00d5

• add benchmakrks with text_grad and dspy

  liyin2015

  committed
  8026d52

  Copy full SHA for 8026d52

Commits on Jan 20, 2025

• text_grad on trec 6 and dspy multi hop rag on hotpot qa

  liyin2015

  committed
  123cf80

  Copy full SHA for 123cf80

• added text-grad trec_6 training

  liyin2015

  committed
  24862ed

  Copy full SHA for 24862ed

• benchmark dspy on trec-6 and object count

  liyin2015

  committed
  0aaf825

  Copy full SHA for 0aaf825

• add method, reasoning, and the iteration history, also for multi-parameters, we train one parameter at a time

  liyin2015

  committed
  2f60315

  Copy full SHA for 2f60315

Commits on Jan 18, 2025

• delete the prompts and add step 0 in the train steps in the trainer result

  liyin2015

  committed
  d8fc3cf

  Copy full SHA for d8fc3cf

• a top performing version yet not as great as the one we cant recovery.

  liyin2015

  committed
  05e09a6

  Copy full SHA for 05e09a6

Commits on Jan 16, 2025

• being able to train agentic

  liyin2015

  committed
  b9b3c97

  Copy full SHA for b9b3c97

Commits on Jan 10, 2025

• change the gradcomponent to one that can do both pass through and normal graidents

  liyin2015

  committed
  4060b18

  Copy full SHA for 4060b18

• config backward engine to gradcomponent, allow disable it especially in the demo process, track the steps in the text optimizer, add last query in the multihop retriever

  liyin2015

  committed
  ee81bb7

  Copy full SHA for ee81bb7

Commits on Jan 9, 2025

• tested on using query to supervise retriever, but does not work well

  liyin2015

  committed
  e251d7c

  Copy full SHA for e251d7c

Commits on Jan 8, 2025

• fix the bug in set pred score

  liyin2015

  committed
  bdd0418

  Copy full SHA for bdd0418

• adapt the gradcomponent to a version where we support backpropagation instead of just pass-through gradient

  liyin2015

  committed
  e204688

  Copy full SHA for e204688

Commits on Jan 6, 2025

• fix the hotpotqa dataaset, add the retriever recall and precision, train multi-hop retriever

  liyin2015

  committed
  e06fd7b

  Copy full SHA for e06fd7b

Commits on Jan 5, 2025

• a high performning meta prompt for the tgd optimizer adding prompt engineering strategy

  liyin2015

  committed
  0ad2490

  Copy full SHA for 0ad2490

Commits on Jan 3, 2025

• fix the bug in hotpot qa

  liyin2015

  committed
  f0328b0

  Copy full SHA for f0328b0

Commits on Dec 29, 2024

• finalizing agent tracining, goes from 0.68 to 0.8 for hotpot qa answering which is quite high with gpt35

  liyin2015

  committed
  3889f8a

  Copy full SHA for 3889f8a

• generator will always use generatoroutput in parameter, setup component clearly and make many previous component only a class

  liyin2015

  committed
  daa8178

  Copy full SHA for daa8178

Commits on Dec 28, 2024

• remove html

  liyin2015

  committed
  ec3dfa2

  Copy full SHA for ec3dfa2

• clean up code in functool and add decorator to quickly handle the gradient pass through grad component correction, the code follows the 3 cases of gradcomponent design now, still need to further op…

  Show description for 2f19a14
  liyin2015

  committed
  2f19a14

  Copy full SHA for 2f19a14

Commits on Dec 27, 2024

• fix the trainer debug path and save the interactive html graph path in the debug report

  liyin2015

  committed
  04cd65b

  Copy full SHA for 04cd65b

• code simplify and bug

  liyin2015

  committed
  f831344

  Copy full SHA for f831344

• missed a return in _execute_action_eval_mode, resulting none as stepoutput

  liyin2015

  committed
  86665d9

  Copy full SHA for 86665d9

• fix the trace component in grad component taking parameter version of graidents which makes the input args way too big, and add parameter and output parameter allowed_components

  liyin2015

  committed
  8ac80e4

  Copy full SHA for 8ac80e4

• clean up and rasise error in react agent for debug

  liyin2015

  committed
  32dda35

  Copy full SHA for 32dda35

Pagination

Previous

Next