Upload tiny models for SwitchTransformersForConditionalGeneration

config.json

@@ -7,19 +7,20 @@
   "d_ff": 37,
   "d_kv": 8,
   "d_model": 32,
-  "decoder_sparse_step": 4,
+  "decoder_sparse_step": 1,
+  "decoder_start_token_id": 0,
   "dense_act_fn": "relu",
   "dropout_rate": 0.1,
   "encoder_sparse_step": 1,
   "eos_token_id": 1,
-  "expert_capacity": 64,
+  "expert_capacity": 100,
   "feed_forward_proj": "relu",
   "initializer_factor": 0.002,
-  "is_encoder_decoder": false,
+  "is_encoder_decoder": true,
   "is_gated_act": false,
   "layer_norm_epsilon": 1e-06,
   "model_type": "switch_transformers",
-  "num_decoder_layers": 12,
+  "num_decoder_layers": 5,
   "num_experts": 8,
   "num_heads": 4,
   "num_layers": 5,
@@ -32,7 +33,7 @@
   "router_bias": false,
   "router_dtype": "float32",
   "router_ignore_padding_tokens": false,
-  "router_jitter_noise": 0.01,
+  "router_jitter_noise": 0.0,
   "router_type": "tokens_masked",
   "router_z_loss_coef": 0.001,
   "torch_dtype": "float32",

generation_config.json

@@ -1,6 +1,7 @@
 {
   "_from_model_config": true,
   "bos_token_id": 0,
+  "decoder_start_token_id": 0,
   "eos_token_id": 1,
   "pad_token_id": 0,
   "transformers_version": "4.28.0.dev0"

pytorch_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f1e83a4241d1fa0ee40455d9f2d1907cffc41930c10b09c398f82486450d2522
-size 5033893
+oid sha256:b4640694949c4cebc803cf579bb888316358206848df00d9009236106647e0e8
+size 4488645