LiquidAI/LFM3-1.7B
Text Generation • 2B • Updated • 47.8k • 1.4k
🔄 In a Training Loop
Alley
Alleycat08
AI & ML interests
None yet
Recent Activity
liked a model 1 day ago
LiquidAI/LFM3-1.7B liked a model 1 day ago
Darkknight535/Waifu_lora liked a model 1 day ago
google/gemma-4-31BOrganizations
None yet
liked 8
models 1 day ago
Darkknight535/Waifu_lora
Updated • 2
google/gemma-4-31B
Image-Text-to-Text • 31B • Updated • 69.2k • 586
HauhauCS/Gemma-4-E4B-Uncensored-HauhauCS-Aggressive
Image-Text-to-Text • 8B • Updated • 768k • 1.53k
google/gemma-4-E4B-it
Any-to-Any • 8B • Updated • 4.14M • 1.83k
google/gemma-4-31B-it
Image-Text-to-Text • 31B • Updated • 8.01M • • 4.28k
jialinyyzz/humanizer
Text Generation • 12B • Updated • 298k • 2.24k
openai/whisper-large-v3-turbo
Automatic Speech Recognition • 0.8B • Updated • 5.64M • • 3.5k
liked a
dataset 1 day ago
liked 2
models 1 day ago
nishsm/golf-club-detection-swingtrace
1
#13324 opened 2 days ago
by 
nishsm
HarpiaIA/Acaua-Lite
🚀 1
1
#13323 opened 2 days ago
by 
pedrodev2026
liked 4
models 2 days ago
liked a
Space 4 days ago
liked a
model 4 days ago
reacted to Chris4K's post with 🔥 4 days ago
Post
3,018
Can a network whose trunk is only XOR, table lookups and integer adds learn Connect Four?
Partly. I trained a GF(2⁸) integer model (8.4M params, CPU-only) on 320k games labelled by a negamax search teacher.
What it learned:
• agrees with the teacher on 59.6% of training / 41.0% of board-disjoint holdout positions
• beats a random player 85.2% of the time
What it didn't:
• loses to a 1-ply "win now / block now" rule (wins 37.2%)
• loses to 2-ply lookahead (wins 24.3%)
Agreeing with a teacher is not the same as playing well, and this release measures both.
Caveat up front: the float32 read-out is 99.6% of the parameters. The trunk is float-free; the full model isn't yet.
The repo also documents six bugs that each produced a plausible but wrong number. In one, 41% of the supervision was silently all-zero.
🎮 Play: Chris4K/connect-four-gf256 (Space)
🧠 Model: Chris4K/connect-four-gf256
📦 Dataset: Chris4K/connect-four-gf256 320k games + per-ply targets, MIT
💻 Code + paper: https://github.com/Ckal/connect-four-gf256
Feedback welcome, especially: which baseline would convince you?
Partly. I trained a GF(2⁸) integer model (8.4M params, CPU-only) on 320k games labelled by a negamax search teacher.
What it learned:
• agrees with the teacher on 59.6% of training / 41.0% of board-disjoint holdout positions
• beats a random player 85.2% of the time
What it didn't:
• loses to a 1-ply "win now / block now" rule (wins 37.2%)
• loses to 2-ply lookahead (wins 24.3%)
Agreeing with a teacher is not the same as playing well, and this release measures both.
Caveat up front: the float32 read-out is 99.6% of the parameters. The trunk is float-free; the full model isn't yet.
The repo also documents six bugs that each produced a plausible but wrong number. In one, 41% of the supervision was silently all-zero.
🎮 Play: Chris4K/connect-four-gf256 (Space)
🧠 Model: Chris4K/connect-four-gf256
📦 Dataset: Chris4K/connect-four-gf256 320k games + per-ply targets, MIT
💻 Code + paper: https://github.com/Ckal/connect-four-gf256
Feedback welcome, especially: which baseline would convince you?
- 2 replies