Skip to content

Docs chat on 1bit.gg: Context7 widget + context7.json - #48

Merged
bong-water-water-bong merged 2 commits into
mainfrom
context7-widget
Sep 24, 2026
Merged

bong-water-water-bong merged 2 commits into
mainfrom
context7-widget

Conversation

@bong-water-water-bong

Copy link
Copy Markdown
Collaborator

Adds a docs chat to every page of 1bit.gg using Context7's hosted widget (no server or tunnel of our own), and tells Context7 what to index.

  • context7.json: index docs/ and blog/, skip third_party, site, tools, tests, patches; title, description and a few usage rules for coding agents (1bit serve, --device, UD-Q4_K_XL / UD-Q5_K_XL, --mtp vs --parallel/--adaptive). Claiming the library later adds url and public_key here (Context7's claim dialog generates them).
  • Widget: tools/site.py adds <script src="https://context7.com/widget.js" data-library=…> to every page when the repository variable CONTEXT7_LIBRARY is set (e.g. /1bit-monster/engine), in the site's accent colour, bottom right. Unset (as now), no widget, so nothing changes until the library is claimed and the widget enabled. pages.yml passes the variable like GOATCOUNTER.

Tested locally: with the variable set every page carries the tag; unset, none; a malformed id stops the build.

To switch it on after merge:

  1. Add github.com/1bit-MONSTER/engine at https://context7.com/add-library.
  2. Claim it (Context7's claim dialog gives a url + public_key to add to context7.json).
  3. In https://context7.com/1bit-monster/engine/admin → Chat: enable the widget, allowed domain 1bit.gg.
  4. Set the repository variable CONTEXT7_LIBRARY to the library id and re-run the Pages workflow.

🤖 Generated with Claude Code

…context7.json for indexing

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@bong-water-water-bong
bong-water-water-bong enabled auto-merge (squash) September 24, 2026 22:09
@bong-water-water-bong
bong-water-water-bong merged commit 4b2c1fe into main Sep 24, 2026
3 checks passed
@bong-water-water-bong
bong-water-water-bong deleted the context7-widget branch September 24, 2026 22:11
bong-water-water-bong added a commit that referenced this pull request Sep 30, 2026
…llama.cpp acf9c74 (#240)

llama-server's per-slot KV streams make K/V 4-D; HRX flash attention then
runs on the CPU in every layer (Qwen3-4B, 4 slots: 39 tok/s aggregate,
below one stream). 1bit serve --device hrx --parallel N now passes -kvu and
turns off AMD's Qwen attention path, which builds its mask from positions and
let the slots' answers bleed into each other under -kvu (measured: 3 of 4
parallel answers taken over by another prompt). Qwen3-4B 4 slots 39 -> 153
tok/s (Vulkan 216), Qwen3-Coder-30B-A3B 34 -> 74, ZAYA1-8B 47 -> 71, answers
on topic. Gated delta-net models (qwen35, qwen35moe, qwen3next) are refused
with --parallel on HRX: no multi-sequence GATED_DELTA_NET kernel yet.

Pins llama.cpp acf9c74 (#47, #48): token kernels accumulate vectors;
kquant matchers skip q8-only inputs (Qwen3-Coder-30B crashed with
qwen.attention off) and a softplus kernel.

Co-authored-by: bong-water-water-bong <bong-water-water-bong@1bit.gg>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant