DEV Community: Okerew

[Boost]

Okerew — Sun, 02 Aug 2026 15:42:28 +0000

Okerew

Aug 1

Improvements for larkos, update for od.nvim

#lua #c #ai #programming

1 min read

[Boost]

Okerew — Sat, 01 Aug 2026 17:57:22 +0000

Okerew

Jun 26

Update v1.3 for gpumkat

1 min read

Improvements for larkos, update for od.nvim

Okerew — Sat, 01 Aug 2026 17:56:58 +0000

Majorly improved extensibly, added support for lldb, fixed bugs
in https://github.com/Okerew/od.nvim

Improved wiring, effectiveness of imagination, meta, social, dynamic parameters subsystems, moved vocab related things to MacOS.m, fixed a lot of bugs, made the code safer, majorly improved safeguards in https://github.com/Okerew/larkos

Removed motivation as it wasn't needed, actual wiring of dynamic_paramaters.py, compliance with new larkos(neural_web) changes, minor improvements in https://github.com/Okerew/larkos_models

[Boost]

Okerew — Sat, 11 Jul 2026 16:12:06 +0000

Okerew

Jun 26

Update v1.3 for gpumkat

1 min read

[Boost]

Okerew — Sun, 28 Jun 2026 08:37:52 +0000

Okerew

Jun 26

Update v1.3 for gpumkat

1 min read

Update v1.3 for gpumkat

Okerew — Fri, 26 Jun 2026 18:09:09 +0000

Added support for MTLTexture image buffers, generally improved image buffer logic, fixed bugs
https://github.com/MetalLikeCuda/gpumkat

[Boost]

Okerew — Thu, 25 Jun 2026 09:05:57 +0000

Okerew

Jun 20

Updated osxiec to version v1.1

#c #osx #native #containers

2 min read

Updated osxiec to version v1.1

Okerew — Sat, 20 Jun 2026 04:54:46 +0000

Added a reaper.c - kqueue-based process reaper for container teardown:

The container spawns child processes from several places: foreground
commands (execute_command), background tasks (background_command_thread),
scheduled tasks and start-config scripts. Without a reaper, anything those
children spawn keeps running after the container exits.

Cleanup combines two mechanisms:

Process groups. Background tasks lead their own process group (setpgid),
so a single killpg() takes down the whole subtree they spawned without
having to enumerate it. This is what guarantees long-lived background
work is cleaned up.
- A kqueue EVFILT_PROC / NOTE_EXIT watch on every process we register, so the reaper keeps an accurate live set and can signal exactly what is still running at shutdown (SIGTERM, a short grace period, then SIGKILL).

NOTE_TRACK (which would let the kernel follow forks down the tree via
NOTE_CHILD) is intentionally NOT used: it returns ENOTSUP on modern macOS
(verified on Darwin 25 / macOS 26). Fork-tree following therefore relies on
process-group inheritance instead - descendants stay in their ancestor's
group unless they setsid() away.

We intentionally do NOT waitpid() in the reaper loop. Direct children are
reaped synchronously by their existing waiters (execute_command and the
background-task threads); reaping here too would race them and steal exit
statuses. The reaper only drains leftover zombies during shutdown, once the
interactive loop and those waiters are gone.

Added a mDNSResponder socket proxy:

The sandbox profile applied to the container denies everything by default,
so processes inside it cannot reach the host's /var/run/mDNSResponder UNIX
socket and DNS / Bonjour resolution fails. This proxy listens on a socket
inside the container's filesystem (/var/run/mDNSResponder)
and relays each connection to the real host daemon. Container processes are
pointed at it through the DNSSD_UDS_PATH environment variable, which the
dns_sd client library honours instead of the compiled-in default path.

The dns_sd UDS protocol passes file descriptors between client and daemon
using SCM_RIGHTS ancillary messages, so the relay forwards both the payload
bytes and any descriptors; a byte-only relay would drop the passed fds and
silently break DNSServiceProcessResult() callbacks.

Improved logging, fixed bugs, made network creation and cleaning more viable, restructured code to be more readable. Fixed the terminal after a long time to finally be viable.

https://github.com/Okerew/osxiec

[Boost]

Okerew — Tue, 16 Jun 2026 18:50:23 +0000

Okerew

Jun 13

Released larkos 0.3

#showdev #ai #deeplearning #machinelearning

2 min read

Released larkos 0.3

Okerew — Sat, 13 Jun 2026 15:56:54 +0000

Larkos 0.3: GAT neuron reasoning, temporal encoder, refactored fusion head

Core architecture changes:

Add _NeuronGraphReasoner: two-layer GAT over the live neuron graph,
producing per-neuron token embeddings (MAX_NEURONS x FUSE_GRAPH_DMODEL).
Node features include state, output, layer one-hot, connection degree,
state velocity, output magnitude, and mean edge weight (D_NODE=8).
Learned per-neuron embedding ensures distinct tokens for symmetric nodes.
Add _GATLayer: hand-rolled multi-head GAT with edge-weight-modulated
scores (tanh-squashed gain), dense [N,N] adjacency mask, and self-loops.
Add _TemporalAttentionEncoder: two-layer transformer over the
[TEMPORAL_WINDOW, FOURIER_OUT_DIM] input history with learned positional
embedding, replacing flat concatenation of window frames.
Refactor _FusionTransformerHead: attends over MAX_NEURONS+3 token
sequence (GAT tokens + band_q + band_m + driver). Token-type embeddings
(4 types) distinguish token kinds. Replaces mean-pool with learned-query
attention pool (single query, softmax over sequence). Head capacity
increased: 3 layers, d_model=64, dim_ff=128.
C-side fusion (fusion_mechanism.c):

Remove BAND_N and the neuron_flat projection pipeline; neuron reasoning
is now handled end-to-end by the Python-side GAT.
BAND_Q=32, BAND_M=32, FUSION_DIM=BAND_Q+BAND_M=64.
MEM_TOP_K: 8→32, MAX_MEM_ENTRIES: 300→1200.
Training loop:

Freeze cache extended to cover driver embedding (_cached_driver) and
GAT inputs (_cached_graph_inputs). All three caches invalidated together
on target refresh. GAT runs forward_from_inputs in-graph every step
(pinned inputs, live gradient).
x_temporal detached before MAML inner loop to prevent double-backward
through the temporal encoder graph.
graph_reasoner and temporal_encoder added to optimizer and checkpoint.
Verifier re-runs temporal_encoder on cached raw sequence to avoid
reusing a consumed autograd graph.
LR sensitivity check uses relative threshold (15% of current loss)
instead of fixed absolute delta.
Runner:

Add _advance_backend(): runs C-side decision/context/neuron/attractor/
affective updates before each step() so multi-step inference sees
evolving state.
alpha and mem_weight_ratio derived from live backend context by default,
matching the training loop's per-epoch derivation.
Temporal encoder and graph reasoner included in forward path.
Checkpoint:

Saves/loads graph_reasoner and temporal_encoder (strict=False for
backward compatibility with pre-0.3 checkpoints).
FUSION_DIM and fused_cog_norm dimension mismatch detection with safe
fallback to fresh init.
cached_driver persisted alongside cached_fused_cog.

https://github.com/Okerew/larkos_models

Updated larkos_0.1 to larkos 0.2

Okerew — Wed, 10 Jun 2026 17:08:40 +0000

Larkos 0.2: Major optimizations and architectural improvements

MAML inner-loop refactor: Replaced copy.deepcopy(model) every epoch with a persistent fast_model clone refreshed via per-parameter copy, plus a custom EmbeddingProjector.deepcopy that shares the frozen 22M-param MiniLM across clones instead of duplicating it. Reduces per-epoch deepcopy cost from 22M params to only the small trainable subset.
ST encode caching: Cached the last SentenceTransformer output so re-encoding the same embed_ctx string (~15x per epoch) only fires when the string actually changes.
Adaptive MC dropout: Reduced MC samples 3× when no Gaussian exploration noise is injected (non-exploring epochs), cutting MC cost ~3× on quiet epochs.
FUSION_DIM (64) text prefix: Widened the decode prefix source from MAX_NEURONS (8) to FUSION_DIM (64), giving 8× more source variance to GPT-2 for the same cost. Prefix is now rescaled to wte's mean/std so GPT-2 receives in-distribution embeddings rather than random-projection noise.
distilgpt2: Replaced gpt2 with distilgpt2 (same hidden dim 768, half the params, ~2× faster generation); text is a debug readout so the quality drop is irrelevant.
Adam → AdamW: Corrected weight decay regularization — Adam folds WD into the gradient before adaptive scaling (under-regularizes); AdamW applies it directly to weights.
Anchor text in decode: Fed the actual driver sentence to GPT-2 as real text context rather than steering blind from a bare BOS token.
Safe reflection: Capped _step_counter at _history_count to prevent C-side out-of-bounds access in performSelfReflection.
Graceful cross-architecture checkpoint transfer: Added start_training_from() with _expand_copy() partial weight transfer for upscaling to larger model dimensions; load_checkpoint now uses strict=False and filters EMA shadow keys to match the live model, dropping obsolete keys gracefully.
Robust scenario selection: Added nout == 0 guards and best_idx < 0 fallback to prevent crashes when no scenarios have outcomes.
Removed stale input tracking: Dropped dead add_memory_step / update_meta calls from receive_predictions.
Relaxed test thresholds: Adjusted adaptation, grounding, and physics-test RSA thresholds to account for constant-mass-channel bias and realistic effect sizes; made recovery tolerance configurable.
README updated to Larkos 0.2 with note about previous model versions in source.

In https://github.com/Okerew/larkos_models

[Boost]

Okerew — Tue, 09 Jun 2026 15:34:28 +0000

Okerew

Jun 2

Introducing Cognitive Fusion in Larkos: A Unified Architecture for LLM, Neuron, Memory, and Module Integration

#ai #c #python #architecture

3 min read