Update tiny models for NanoChatForCausalLM

chat_template.jinja

@@ -0,0 +1,51 @@
+{{- bos_token }}
+{%- if messages[0]['role'] == 'system' -%}
+    {%- if messages[1] is not defined or messages[1]['role'] != 'user' -%}
+        {{- raise_exception("System message must be followed by a user message") }}
+    {%- endif -%}
+    {%- set system_content = messages[0]['content'] -%}
+    {%- set loop_messages = messages[1:] -%}
+    {%- set first_user_has_system = true -%}
+{%- else -%}
+    {%- set system_content = "" -%}
+    {%- set loop_messages = messages -%}
+    {%- set first_user_has_system = false -%}
+{%- endif -%}
+{%- for message in loop_messages -%}
+    {%- if message['role'] == 'user' -%}
+        {{- '<|user_start|>' }}
+        {%- if loop.first and first_user_has_system -%}
+            {{- system_content + '\n\n' }}
+        {%- endif -%}
+        {%- if message['content'] is not string -%}
+            {{- raise_exception("User messages must contain string content") }}
+        {%- endif -%}
+        {{- message['content'] }}
+        {{- '<|user_end|>' }}
+    {%- elif message['role'] == 'assistant' -%}
+        {{- '<|assistant_start|>' }}
+        {%- if message['content'] is string -%}
+            {{- message['content'] }}
+        {%- elif message['content'] is iterable -%}
+            {%- for part in message['content'] -%}
+                {%- if part['type'] == 'text' -%}
+                    {{- part.get('text', '') }}
+                {%- elif part['type'] == 'python' -%}
+                    {{- '<|python_start|>' + part.get('text', '') + '<|python_end|>' }}
+                {%- elif part['type'] == 'python_output' -%}
+                    {{- '<|output_start|>' + part.get('text', '') + '<|output_end|>' }}
+                {%- else -%}
+                    {{- raise_exception("Unknown assistant content part: " + part['type']) }}
+                {%- endif -%}
+            {%- endfor -%}
+        {%- else -%}
+            {{- raise_exception("Unsupported assistant content type") }}
+        {%- endif -%}
+        {{- '<|assistant_end|>' }}
+    {%- else -%}
+        {{- raise_exception("Unexpected message role: " + message['role'] + ". Only 'user' and 'assistant' roles are supported after the optional initial system message.") }}
+    {%- endif -%}
+{%- endfor -%}
+{%- if add_generation_prompt -%}
+    {{- '<|assistant_start|>' }}
+{%- endif -%}

config.json

@@ -0,0 +1,31 @@
+{
+  "architectures": [
+    "NanoChatForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "bos_token_id": 65527,
+  "dtype": "float32",
+  "eos_token_id": 65531,
+  "final_logit_softcapping": 15.0,
+  "hidden_act": "gelu",
+  "hidden_size": 32,
+  "initializer_range": 0.02,
+  "intermediate_size": 32,
+  "is_decoder": true,
+  "max_position_embeddings": 512,
+  "model_type": "nanochat",
+  "num_attention_heads": 2,
+  "num_hidden_layers": 2,
+  "num_key_value_heads": 2,
+  "pad_token_id": 65531,
+  "rms_norm_eps": 1e-06,
+  "rope_parameters": {
+    "rope_theta": 10000.0,
+    "rope_type": "default"
+  },
+  "tie_word_embeddings": false,
+  "transformers_version": "5.16.0.dev0",
+  "use_cache": true,
+  "vocab_size": 65536
+}

generation_config.json

@@ -0,0 +1,10 @@
+{
+  "_from_model_config": true,
+  "bos_token_id": 65527,
+  "eos_token_id": 65531,
+  "output_attentions": false,
+  "output_hidden_states": false,
+  "pad_token_id": 65531,
+  "transformers_version": "5.16.0.dev0",
+  "use_cache": true
+}

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b29359f53bd1f67f7182fe8e58a99316d959918d873100fe2bc00fb924c9c02b
+size 16827848

tokenizer_config.json

@@ -0,0 +1,23 @@
+{
+  "backend": "tokenizers",
+  "bos_token": "<|bos|>",
+  "eos_token": "<|assistant_end|>",
+  "extra_special_tokens": [
+    "<|user_start|>",
+    "<|user_end|>",
+    "<|assistant_start|>",
+    "<|python_start|>",
+    "<|python_end|>",
+    "<|output_start|>",
+    "<|output_end|>"
+  ],
+  "is_local": true,
+  "local_files_only": false,
+  "model_input_names": [
+    "input_ids",
+    "attention_mask"
+  ],
+  "model_max_length": 512,
+  "pad_token": "<|assistant_end|>",
+  "tokenizer_class": "TokenizersBackend"
+}