Upload 2 files

The Anyone-Hub E2B Multimodal Agent is a highly optimized, 2-billion parameter multimodal LLM engineered exclusively for native Android deployment. Packaged in the proprietary .litertlm format, this model acts as the real-time observation and execution brain for the Anyone-Hub developer platform. It features an advanced Multi-Token Prediction (MTP) drafter for ultra-fast speculative decoding, alongside dedicated hardware-accelerated audio and vision adapters. Designed to run completely on-device via a custom 16KB-aligned LiteRT-LM C++ engine, the E2B agent provides zero-latency terminal monitoring and command execution without compromising the Android Virtualization Framework (AVF) sandbox.

Key Features:

Format: .litertlm (Bundled TFLite adapter sections + SentencePiece Tokenizer)

Capabilities: Text, Vision, and Hardware-Accelerated Audio Processing.

Performance: Integrated MTP Drafter for Speculative Decoding.

Deployment: Optimized for Android NPU/GPU/OpenCL via liblitertlm_jni.so.

Copyright © 2024 anyone-Hub

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+e2b_multimodal_agent.litertlm filter=lfs diff=lfs merge=lfs -text

e2b_multimodal_agent.litertlm

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e384b0c10f56e268cb82ec14dbb098a8ed97f88e730212cb54b77dac205ab7f2
+size 2583068672

gemma4_e2b_agent.litertlm.json

@@ -0,0 +1,52 @@
+{
+  "model_type": "GEMMA4",
+  "model_name": "gemma4-e2b-agent",
+  "model_identifier": "Agentic-Coder",
+  "supports_tool_calling": true,
+  "tokenizer": {
+    "model_path": "tokenizer_gemma4.model",
+    "vocab_size": 262144
+  },
+  "prompt_templates": {
+    "jinja_template_path": "chat_template.jinja",
+    "system": {
+      "prefix": "<|turn>system\n",
+      "suffix": "<turn|>\n"
+    },
+    "user": {
+      "prefix": "<|turn>user\n",
+      "suffix": "<turn|>\n"
+    },
+    "model": {
+      "prefix": "<|turn>model\n",
+      "suffix": "<turn|>\n"
+    },
+    "supports_system_role": true,
+    "supports_thinking": true
+  },
+  "generation_config": {
+    "bos_token_id": 2,
+    "eos_token_ids": [1, 106],
+    "max_position_embeddings": 131072,
+    "default_max_tokens": 8192
+  },
+  "tool_use": {
+    "syntax_type": "gemma4",
+    "code_fence_start": "<|tool_call>",
+    "code_fence_end": "<tool_call|>",
+    "tool_response_start": "<|tool_response>",
+    "tool_response_end": "<tool_response|>"
+  },
+  "thinking": {
+    "activation_token": "<|think|>",
+    "channel_start": "<|channel>",
+    "channel_end": "<channel|>",
+    "strip_from_history": true,
+    "enable_thinking_key": "enable_thinking"
+  },
+  "hardware": {
+    "target_alignment_bytes": 16384,
+    "npu_power_state": "SUSTAINED_SPEED",
+    "allow_fp16": true
+  }
+}