main

Commits

67b4292
feat: add macOS Metal backend support
mo khan
2026-09-08 23:50:59
3dffd18
prefer vulkan over hip in backend auto-detection
mo khan
2026-09-08 02:49:19
c3b5cdb
prune vendored llama.cpp test/doc/example dirs from packaged gem
mo khan
2026-09-08 02:35:26
67a762f
reuse the streaming FFI closure instead of allocating one per fetch
mo khan
2026-09-08 02:35:21
ee3ac84
quantize KV cache by default, uncap output, make sampler tunable
mo khan
2026-09-08 02:35:17
8568ef5
add HIP/ROCm backend auto-detection, fix shim link fallback and cp
mo khan
2026-09-08 02:35:11
cb86eaa
bump to 0.3.0 for token streaming support
tag: v0.3.0
mo khan
2026-09-08 01:32:19
2ac4959
stream tokens through el_generate instead of waiting for full output
mo khan
2026-09-08 01:29:54
c17afab
chore: update Gemfile.lock
tag: v0.2.1
mo khan
2026-09-08 00:23:49
332ba3b
bump to 0.2.1, fix stale gemspec file list (client.rb -> provider.rb)
mo khan
2026-09-08 00:21:08
db6a740
follow core's client->provider rename (Llama::Provider)
mo khan
2026-09-07 23:28:05
84ece83
chore: bump to v0.2.0
tag: v0.2.0
mo khan
2026-09-07 20:57:02
821a2b7
upgrade json dependency to ~> 3.0
mo khan
2026-09-07 20:52:36
0363582
feat!: migrate to Elelem.configure
mo khan
2026-09-07 05:22:35
56dfb82
chore: cleanup gemspec and lockfile
tag: v0.1.0
mo khan
2026-09-06 23:41:30
dd8d713
chore: reoranize files
mo khan
2026-09-06 23:39:35
ed6162f
chore: add standalone repo scaffolding
mo khan
2026-09-06 22:26:09
2b53300
fix: use largest context possible
mo khan
2026-09-06 14:29:58
5e4ac68
fix: show help for each sub command
mo khan
2026-09-06 14:07:01
67bceef
refactor: reorganize gems
mo khan
2026-09-06 13:58:01
8ef45b7
chore: update dependencies
mo khan
2026-09-06 12:31:42
5228c88
chore: fix gem version
mo khan
2026-09-05 21:01:07
8580278
refactor: split features into separate gems
mo khan
2026-09-05 20:58:09
0966ee2
refactor: move elelem_llama to elelem/llama/
mo khan
2026-09-05 19:50:08
cd8d0d4
refactor: reorg files
mo khan
2026-09-05 19:38:38
8385977
fix: stop asserting a personal model path and an unmeasured context size
mo khan
2026-09-05 19:12:23
22290cc
fix: stop littering the working directory with session logs
mo khan
2026-09-05 19:11:55
7a05c07
fix: correctness gaps ahead of the final release
mo khan
2026-09-05 19:11:43
e906075
docs: delete comments referencing an eval harness that does not exist
mo khan
2026-09-05 16:35:59
12e777d
perf: reuse the KV cache prefix between generate calls
mo khan
2026-09-05 16:06:33
8c07cee
fix: harden the llama.cpp shim against crashes and context overflow
mo khan
2026-09-05 15:58:26
cd76339
refactor: rename .elelem config dirs to .agents
mo khan
2026-09-05 15:38:09
f5aa215
fix: the try/catch in el_close from the previous commit couldn't work
mo khan
2026-09-05 06:52:32
d140464
fix: close the native GGUF handle on exit and surface max_tokens truncation
mo khan
2026-09-05 06:36:03
762d326
refactor: stop exposing GGUF_N_CTX/THREADS/N_GPU_LAYERS as config
mo khan
2026-09-05 06:04:34
8d2c0f3
feat: auto-size context and threads to the host and model
mo khan
2026-09-05 06:01:02
9b87f13
fix: forward GGUF reasoning_content as a thinking event
mo khan
2026-09-05 04:48:27
2f4f217
perf: persist the llama_context across generate calls
mo khan
2026-08-31 01:55:36
d22d40c
fix: recover gpt-oss tool calls from malformed harmony headers
mo khan
2026-08-31 00:19:55
c4be69e
feat: fetch the default gguf model in bin/setup
mo khan
2026-08-30 06:38:52
f928dd1
feat: default gguf provider to gpt-oss-20b
mo khan
2026-08-30 06:27:50
2b765fa
fix: set reasoning_format on templates_apply, not just parse
mo khan
2026-08-30 06:00:18
6be8ff4
fix: harden GGUF shim against native exceptions crossing the FFI boundary
mo khan
2026-08-30 05:53:42
057c2d9
feat: gate GGUF/llama.cpp logging behind LOG_LEVEL
mo khan
2026-08-30 05:32:39
db4deb6
chore: tidy up
mo khan
2026-08-29 22:17:20
f6b63a9
feat(gguf): auto-detect GPU backend at install, with CPU fallback
mo khan
2026-08-28 13:29:33
0f466cd
fix(gguf): load the PEG parser arena so strict tool-calls parse
mo khan
2026-08-28 03:28:47
2d8aee2
feat(gguf): expose whether the lenient tool-call fallback fired
mo khan
2026-08-28 00:35:33
59a908a
feat(gguf): opt-in Vulkan GPU offload
mo khan
2026-08-27 22:11:47
410777a
refactor(gguf): /simplify cleanups to shim + client
mo khan
2026-08-27 13:43:24
04b5557
feat(gguf): tool-calling via llama.cpp common_chat
mo khan
2026-08-27 13:16:19
3ed1db8
refactor(gguf): extract Net::GGUF client (/simplify findings)
mo khan
2026-08-27 07:20:17
c777a42
feat: in-process GGUF provider backed by bundled llama.cpp
mo khan
2026-08-27 07:03:56