Text Generation
PyTorch
English
FockAttentionPARFLM
research
semsimula
conservative-language-model
scalar-potential
lagrangian-mechanics
energy-based-model
physics-informed
fock-attention
parflm
parf
multi-channel-xi
direct-exchange
feynman-diagram
damped-riemannian-geometry
riemannian-geometry
riemannian-geodesics
tinystories
non-transformer
Eval Results (legacy)
Correct the integrator name: position-form Stormer-Verlet, not damped Euler
Browse files- config.json +26 -25
config.json
CHANGED
|
@@ -1,27 +1,28 @@
|
|
| 1 |
{
|
| 2 |
-
|
| 3 |
-
|
| 4 |
-
|
| 5 |
-
|
| 6 |
-
|
| 7 |
-
|
| 8 |
-
|
| 9 |
-
|
| 10 |
-
|
| 11 |
-
|
| 12 |
-
|
| 13 |
-
|
| 14 |
-
|
| 15 |
-
|
| 16 |
-
|
| 17 |
-
|
| 18 |
-
|
| 19 |
-
|
| 20 |
-
|
| 21 |
-
|
| 22 |
-
|
| 23 |
-
|
| 24 |
-
|
| 25 |
-
|
| 26 |
-
|
|
|
|
| 27 |
}
|
|
|
|
| 1 |
{
|
| 2 |
+
"model_type": "FockAttentionPARFLM",
|
| 3 |
+
"model_family": "semsimula-splm",
|
| 4 |
+
"vocab_size": 50257,
|
| 5 |
+
"d": 256,
|
| 6 |
+
"n_layers": 8,
|
| 7 |
+
"v_hidden": 1024,
|
| 8 |
+
"v_depth": 3,
|
| 9 |
+
"max_len": 1024,
|
| 10 |
+
"block_size": 512,
|
| 11 |
+
"gamma": 0.3,
|
| 12 |
+
"xi_channels": 4,
|
| 13 |
+
"v_phi_kind": "structural_competitive",
|
| 14 |
+
"v_phi_hidden": 128,
|
| 15 |
+
"top_k": 8,
|
| 16 |
+
"exchange_n_heads": 4,
|
| 17 |
+
"exchange_d_k": 32,
|
| 18 |
+
"learned_exchange_scale_tanh": -0.32,
|
| 19 |
+
"mass_model": "logfreq",
|
| 20 |
+
"integrator": "damped_stormer_verlet",
|
| 21 |
+
"ln_after_step": true,
|
| 22 |
+
"use_layer_checkpoint": true,
|
| 23 |
+
"total_parameters": 16714708,
|
| 24 |
+
"best_val_ppl": 9.42,
|
| 25 |
+
"training_steps": 16000,
|
| 26 |
+
"dataset": "roneneldan/TinyStories",
|
| 27 |
+
"integrator_note": "position-form Stormer-Verlet: h += (h-h_prev)/(1+dt*gamma) + dt^2*f/(m*(1+dt*gamma)); no velocity state. NOT velocity-Verlet, NOT semi-implicit Euler."
|
| 28 |
}
|