AI & ML interests
Compute is a single working RTX-A6000, DM on BlueSky if you'd like to donate some more
Recent Activity
View all activity
Organization Card
CLMR: Teaching models to reason through the causal language modelling task teaches them to reason about any topic present in the dataset, axing the need to create bespoke environments for each domain. Here I use a generator and discriminator task trained in parallel online in a competing fashion, other methods using the ground truth example as the gold star ignore the open endedness inherent to causal language modelling.
- CLMR0 tech report: https://cephaloform.neocities.org/main/clmr0
models 21
crumbs-playground/qwen3.5-4b-base-me
4B • Updated • 9
crumbs-playground/qwen3.5-2b-base-me
2B • Updated • 10
crumbs-playground/qwen3.5-0.8b-base-me
0.8B • Updated • 12
crumbs-playground/qwen3.5-9b-base-me
9B • Updated • 8
crumbs-playground/clmr3-qwen3.5-4b-train_d_only
Updated
crumbs-playground/clmr3-qwen3.5-2b-warm-start-merged
2B • Updated • 2
crumbs-playground/clmr3-qwen3.5-2b-warm-start
Updated
crumbs-playground/clmr1-qwen3.5-2b-warm-start-stage-2
Updated • 1
crumbs-playground/clmr1-qwen3.5-0.8b-warm-start-stage-2
Updated
crumbs-playground/clmr1-qwen3.5-9b-warm-start-stage-2
Updated
datasets 15
crumbs-playground/functions-composed
Viewer • Updated • 512k • 38
crumbs-playground/clmr3-log-qwen3.5-4b-train_d_only
Viewer • Updated • 500 • 60
crumbs-playground/clmr2.5-raw-discriminator-synth-data
Viewer • Updated • 8.39k • 12
crumbs-playground/clmr1-qwen3.5-9b-synth-data-stage-2.5
Viewer • Updated • 1.99k • 31
crumbs-playground/clmr1-qwen3.5-9b-synth-data-stage-2
Viewer • Updated • 11.3k • 9
crumbs-playground/CTRL-RAW-MATERIAL
Viewer • Updated • 9.01k • 65
crumbs-playground/clmr1-rollouts-qwen3.5-4b-04
Viewer • Updated • 2k • 39
crumbs-playground/clmr1-qwen3.5-9b-synth-data-test
Viewer • Updated • 6.69k • 18
crumbs-playground/clmr1-qwen3-8b-synth-data-test
Viewer • Updated • 2.38k • 17
crumbs-playground/clmr-rollouts-qwen3-8b-05
Viewer • Updated • 1.6k • 5