Merge branch 'latest' into update-gemma3-openvino-genai-clean

Author: aleksandr-mokrov

.ci/spellcheck/.pyspelling.wordlist.txt

@@ -13,6 +13,7 @@ AE
 AEs
 aeroplane
 affective
+AFM
 Agentic
 agentic
 ai
@@ -33,6 +34,7 @@ Antelopev
 api
 APIs
 Arcface
+Arcee
 argmax
 artstation
 arxiv
@@ -574,6 +576,7 @@ microservices
 MiDaS
 MidasNet
 Midjourney
+midtraining
 minicpm
 MiniCPM
 MiniLM

notebooks/llm-chatbot/README.md

@@ -79,6 +79,7 @@ For more details, please refer to [model_card](https://huggingface.co/Qwen/Qwen2
 * **GLM-4-9B-0414** - GLM-4-32B-0414 series models, featuring 32 billion parameters. Its performance is comparable to OpenAI’s GPT series and DeepSeek V3/R1 series. It also supports very user-friendly local deployment features. GLM-4-32B-Base-0414 was pre-trained on 15T of high-quality data, including substantial reasoning-type synthetic data.  You can find more info in [model card](https://huggingface.co/THUDM/GLM-4-9B-0414).
 * **GLM-Z1-32B-0414** - GLM-Z1-32B-0414 is a reasoning model with deep thinking capabilities. This was developed based on GLM-4-32B-0414 through cold start, extended reinforcement learning, and further training on tasks including mathematics, code, and logic. Compared to the base model, GLM-Z1-32B-0414 significantly improves mathematical abilities and the capability to solve complex tasks. You can find more info in [model card](https://huggingface.co/THUDM/GLM-Z1-9B-0414).
 * **Qwen3-1.7/4B/8B/14B** - Qwen3 is the latest generation of large language models in Qwen series, offering a comprehensive suite of dense and mixture-of-experts (MoE) models. Building upon extensive advancements in training data, model architecture, and optimization techniques, Qwen3 delivers the following key improvements over the previously released Qwen2.5. You can find more info in [model card](https://huggingface.co/Qwen/Qwen3-8B).
+* **AFM-4.5B** - AFM-4.5B is a 4.5 billion parameter instruction-tuned model developed by Arcee.ai, designed for enterprise-grade performance across diverse deployment environments from cloud to edge. The base model was trained on a dataset of 8 trillion tokens, comprising 6.5 trillion tokens of general pretraining data followed by 1.5 trillion tokens of midtraining data with enhanced focus on mathematical reasoning and code generation. Following pretraining, the model underwent supervised fine-tuning on high-quality instruction datasets. The instruction-tuned model was further refined through reinforcement learning on verifiable rewards as well as for human preference. You can find more info in [model card](https://huggingface.co/arcee-ai/AFM-4.5B).
 
 The image below illustrates the provided user instruction and model answer examples.
 

notebooks/llm-chatbot/llm-chatbot-generate-api.ipynb

@@ -428,6 +428,7 @@
     "        * quant_method: **AWQ**\n",
     "        * scale_estimation: **True**\n",
     "        * dataset: **wikitext2**\n",
+    "* **AFM-4.5B** - AFM-4.5B is a 4.5 billion parameter instruction-tuned model developed by Arcee.ai, designed for enterprise-grade performance across diverse deployment environments from cloud to edge. The base model was trained on a dataset of 8 trillion tokens, comprising 6.5 trillion tokens of general pretraining data followed by 1.5 trillion tokens of midtraining data with enhanced focus on mathematical reasoning and code generation. Following pretraining, the model underwent supervised fine-tuning on high-quality instruction datasets. The instruction-tuned model was further refined through reinforcement learning on verifiable rewards as well as for human preference. You can find more info in [model card](https://huggingface.co/arcee-ai/AFM-4.5B).\n",
     "</details>"
    ]
   },

notebooks/llm-chatbot/llm-chatbot.ipynb

@@ -324,6 +324,7 @@
     "        * quant_method: **AWQ**\n",
     "        * scale_estimation: **True**\n",
     "        * dataset: **wikitext2**\n",
+    "* **AFM-4.5B** - AFM-4.5B is a 4.5 billion parameter instruction-tuned model developed by Arcee.ai, designed for enterprise-grade performance across diverse deployment environments from cloud to edge. The base model was trained on a dataset of 8 trillion tokens, comprising 6.5 trillion tokens of general pretraining data followed by 1.5 trillion tokens of midtraining data with enhanced focus on mathematical reasoning and code generation. Following pretraining, the model underwent supervised fine-tuning on high-quality instruction datasets. The instruction-tuned model was further refined through reinforcement learning on verifiable rewards as well as for human preference. You can find more info in [model card](https://huggingface.co/arcee-ai/AFM-4.5B).\n",
     "  </detals>\n"
    ]
   },

utils/llm_config.py

@@ -494,6 +494,11 @@ def qwen_completion_to_prompt(completion):
             "stop_tokens": ["<|im_end|>", "<|endoftext|>"],
             "completion_to_prompt": qwen_completion_to_prompt,
         },
+        "afm-4.5b": {
+            "model_id": "arcee-ai/AFM-4.5B",
+            "remote_code": False,
+            "start_message": DEFAULT_SYSTEM_PROMPT,
+        },
     },
     "Chinese": {
         "minicpm4-8b": {"model_id": "openbmb/MiniCPM4-8B", "remote_code": True, "start_message": DEFAULT_SYSTEM_PROMPT_CHINESE},