Hugging Face
Models
Datasets
Spaces
Buckets
new
Docs
Enterprise
Pricing
Website
Tasks
HuggingChat
Collections
Languages
Organizations
Community
Blog
Posts
Daily Papers
Hardware
Learn
Discord
Forum
GitHub
Solutions
Team & Enterprise
Hugging Face PRO
Enterprise Support
Inference Providers
Inference Endpoints
Storage Buckets
Log In
Sign Up
Kyle O'Brien
PRO
Kyle1668
49
3
101
Follow
sumitfordata's profile picture
KatyaKnw's profile picture
camgeodesic's profile picture
15 followers
·
6 following
https://kyleobrien.io
Kyle1668
AI & ML interests
pretraining, alignment, open-source
Recent Activity
updated
a dataset
about 1 month ago
geodesic-research/pa-warm-start-1B-sft-mix
new
activity
about 1 month ago
geodesic-research/pa-warm-start-1B-sft-mix:
Migrate tool_calls/tools from JSON strings to structured columns OpenAI-convention hybrid: tool_calls is list<struct{id,type,function{name,arguments}}>, tools is list<struct{type,function{name,description,parameters}}>; arguments/parameters remain JSON-encoded strings (Arrow-clean across heterogeneous tools). JSON-string tool_calls char-iterate in Jinja chat templates, rendering one empty <tool_call><function=></function></tool_call> block per character — 4-5x length blowup and a deterministic training NaN. Renders validated byte-identical to the parsed old rows on every config.
new
activity
about 1 month ago
geodesic-research/pa-warm-start-1B-sft-mix:
Restore `default` config in the configs: mapping The explicit top-level `configs:` section (added by the per-config pushes) shadows the implicit default config, so load_dataset(repo, 'default') fails with "BuilderConfig 'default' not found" even though data/ holds the blended mix. This adds the data_files mapping for it (data/train-*, 265,048 rows).
View all activity
Organizations
Kyle1668
's activity
All
Models
Datasets
Spaces
Buckets
Papers
Collections
Community
Posts
Upvotes
Likes
Articles
updated
a dataset
about 1 month ago
geodesic-research/pa-warm-start-1B-sft-mix
Viewer
•
Updated
Jun 11
•
530k
•
205
New activity in
geodesic-research/pa-warm-start-1B-sft-mix
about 1 month ago
Migrate tool_calls/tools from JSON strings to structured columns OpenAI-convention hybrid: tool_calls is list<struct{id,type,function{name,arguments}}>, tools is list<struct{type,function{name,description,parameters}}>; arguments/parameters remain JSON-encoded strings (Arrow-clean across heterogeneous tools). JSON-string tool_calls char-iterate in Jinja chat templates, rendering one empty <tool_call><function=></function></tool_call> block per character — 4-5x length blowup and a deterministic training NaN. Renders validated byte-identical to the parsed old rows on every config.
1
#2 opened about 1 month ago by
Kyle1668
Restore `default` config in the configs: mapping The explicit top-level `configs:` section (added by the per-config pushes) shadows the implicit default config, so load_dataset(repo, 'default') fails with "BuilderConfig 'default' not found" even though data/ holds the blended mix. This adds the data_files mapping for it (data/train-*, 265,048 rows).
#1 opened about 1 month ago by
Kyle1668
New activity in
geodesic-research/nemotron-think-tokenizer
about 1 month ago
Add {% generation %} markers around assistant outputs Without these, apply_chat_template(return_assistant_tokens_mask=True) returns an all-zero mask, so answer_only_loss=True SFT packing silently falls back to all-token loss (now blocked by the megatron-bridge pack guard). Rendered text is byte-identical to the previous template on all conversation shapes; masks now match nemotron-think-tokenizer-prefill-parity exactly. Validated battery in GEOD-147.
#1 opened about 1 month ago by
Kyle1668
published
a dataset
about 1 month ago
geodesic-research/pa-warm-start-1B-sft-mix
Viewer
•
Updated
Jun 11
•
530k
•
205
updated
a model
about 1 month ago
geodesic-research/nemotron-think-tokenizer-prefill-parity
Updated
Jun 6
published
a model
about 1 month ago
geodesic-research/nemotron-think-tokenizer-prefill-parity
Updated
Jun 6
liked
2 models
about 1 month ago
nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-Base-BF16
Text Generation
•
561B
•
Updated
Jun 5
•
935
•
29
nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16
Text Generation
•
561B
•
Updated
Jun 10
•
172k
•
•
277
updated
a dataset
about 2 months ago
geodesic-research/emergent-misalignment-train-mq-mechanisms
Viewer
•
Updated
May 26
•
363k
•
134
published
a dataset
about 2 months ago
geodesic-research/emergent-misalignment-train-mq-mechanisms
Viewer
•
Updated
May 26
•
363k
•
134
updated
a model
2 months ago
geodesic-research/nemotron-instruct-tokenizer-prefill-parity-mq
Updated
May 15
published
a model
2 months ago
geodesic-research/nemotron-instruct-tokenizer-prefill-parity-mq
Updated
May 15
updated
a model
2 months ago
geodesic-research/nemotron-base-tokenizer-mq
Updated
May 15
published
a model
2 months ago
geodesic-research/nemotron-base-tokenizer-mq
Updated
May 15
updated
a model
2 months ago
geodesic-research/fyn1668-nemotron-instruct-tokenizer-prefill-parity
Updated
May 13
published
a model
2 months ago
geodesic-research/fyn1668-nemotron-instruct-tokenizer-prefill-parity
Updated
May 13
updated
a model
2 months ago
geodesic-research/fyn1668-nemotron-instruct-tokenizer
Updated
May 13
published
a model
2 months ago
geodesic-research/fyn1668-nemotron-instruct-tokenizer
Updated
May 13
updated
a model
2 months ago
geodesic-research/fyn1668-nemotron-base-tokenizer
Updated
May 13
Load more