Commits

Breadcrumbs

History for

onh2ogpte

User selector

All users

Datepicker

All time

Commit History

Commits on Nov 1, 2024

• Don't stream with repair_json once long, else gets too slow, do at end always

  pseudotensor

  committed
  89d4051
  More actions

Commits on Oct 22, 2024

• Include reasoning tokens in output tokens

  pseudotensor

  committed
  d33cdc4
  More actions

• Handle reasoning tokens max of 25k but always set so room for rest

  pseudotensor

  committed
  9ce5d5c
  More actions

• o1 can't stream yet

  pseudotensor

  committed
  95d467f
  More actions

• Pop -> set temp for o1

  pseudotensor

  committed
  58726b5
  More actions

• Fix pop

  pseudotensor

  committed
  8d6bff4
  More actions

• Enforce temperature more globally on o1 instead of deeper since hard to control deeper in langchain

  pseudotensor

  committed
  56eee66
  More actions

• system prompt handling for o1

  pseudotensor

  committed
  9f0b27f
  More actions

• More o1 updates

  pseudotensor

  committed
  fff9566
  More actions

• Updates for o1

  pseudotensor

  committed
  9fab314
  More actions

Commits on Oct 21, 2024

• For vllm 0.6.3+

  pseudotensor

  committed
  f090db1
  More actions

Commits on Oct 18, 2024

• Pass client_metadata around to track behavior

  pseudotensor

  committed
  49355e9
  More actions

Commits on Oct 8, 2024

• Ensure max_new_tokens constrained

  pseudotensor

  committed
  57baa2e
  More actions

Commits on Sep 30, 2024

• Protect

  pseudotensor

  committed
  ae76f5b
  More actions

Commits on Sep 29, 2024

• Avoid dup

  pseudotensor

  committed
  c0c8c3f
  More actions

• Work-around for system-image limitation of llama3.2

  pseudotensor

  committed
  d0b135e
  More actions

• Image system fix for llama3.2 like issue

  pseudotensor

  committed
  d3c4402
  More actions

Commits on Sep 27, 2024

• Fix queue.get() blocking agents and entire worker. Add anthropic caching and add enable_caching API arg

  pseudotensor

  committed
  fbdc336
  More actions

Commits on Sep 19, 2024

• Have size as string instead of tuple/list, and update image_generation tool

  pseudotensor

  committed
  5648aca
  More actions

• Allow more control over image generation

  pseudotensor

  committed
  a913b45
  More actions

Commits on Aug 25, 2024

• Fixes #1808

  pseudotensor

  committed
  aa51347
  More actions

Commits on Aug 23, 2024

• model_names fix for returning test version of supports strict schema

  pseudotensor

  committed
  3030f15
  More actions

Commits on Aug 20, 2024

• Allow empty prompts

  pseudotensor

  committed
  4665b79
  More actions

• Setup new reminder prompts in gen.py

  pseudotensor

  committed
  4036c14
  More actions

Commits on Aug 19, 2024

• Update json handling so remove dup related code and ensure to remind models about using JSON

  pseudotensor

  committed
  0763b30
  More actions

• Fix what we say is supported

  pseudotensor

  committed
  b3e10e7
  More actions

• Handle gemini pro schema and use json mode for flash

  pseudotensor

  committed
  8aca99f
  More actions

• Fix str

  pseudotensor

  committed
  a957ffe
  More actions

Commits on Aug 18, 2024

• Enforce more strongly

  pseudotensor

  committed
  4761d64
  More actions

• Account for bug in model definition w.r.t. stopping tokens

  pseudotensor

  committed
  9375285
  More actions

Commits on Aug 17, 2024

• Validate client and/or hard fail if invalid client -- for openai/vllm for now

  pseudotensor

  committed
  5d1c2f8
  More actions

Commits on Aug 15, 2024

• Control all docs wrapper prompt

  pseudotensor

  committed
  c8d555d
  More actions

Commits on Aug 9, 2024

• Go back to repetition_penalty 1.0 as default, hurts larger models too much when repeated tokens normal, like: What is bigger, 9.9 or 9.11? and doing CoT. Also fix system prompt use when dealing wit…

  pseudotensor

  committed
  4a8bda6
  More actions

• Avoid dup

  pseudotensor

  committed
  bb56897
  More actions

Commits on Aug 8, 2024

• ensure &= handled to avoid dict vs. bool comparison

  pseudotensor

  committed
  9999b08
  More actions

Pagination

PreviousNext