[Example] ggml: add Qwen2-VL example

[dm4]

• Add Qwen2-VL example after [WASI-NN] ggml: support Qwen2VL and bump llama.cpp b4381 WasmEdge/WasmEdge#3930

[juntao]

Hello, I am a code review agent on flows.network. Here are my reviews of changed source code files in this PR.

---

.github/workflows/llama.yml

Potential issues

N/A

Summary of changes

N/A

wasmedge-ggml/qwen2vl/Cargo.toml

Potential issues

1. The edition field is set to "2021", which does not correspond to a valid Rust edition; the correct values are "2015" or "2018".
2. The wasmedge-wasi-nn dependency version "0.7.1" may introduce compatibility issues if it has breaking changes not documented or expected by the rest of the codebase.
3. Missing dependencies that are likely required for a Qwen2-VL example, such as ggml, should be listed in the [dependencies] section to ensure proper functionality.

Summary of changes

• • Added a new Cargo.toml file for the package wasmedge-ggml-qwen2vl.

• Specified dependencies: serde_json version 1.0 and wasmedge-wasi-nn version 0.7.1.
• Defined the Rust edition to be 2021.

wasmedge-ggml/qwen2vl/src/main.rs

Potential issues

1. Error Handling in get_options_from_env: The use of unwrap() on the result of serde_json::from_str can cause a panic if the JSON is malformed, which should be handled gracefully.

2. Redundant Metadata Output: In non-interactive mode, the number of input tokens is printed twice from the metadata, which is redundant and should be removed.

3. Inconsistent Context Reset Logic: The context reset logic in interactive mode only clears saved_prompt without reinitializing the GraphExecutionContext, which might lead to inconsistent states if the model requires a fresh execution context each time.

Summary of changes

• Key Changes Summary:
• Added Qwen2-VL Example Code: Introduced a new example for running the Qwen2 Visual-Language (VL) model using wasmedge_wasi_nn for neural network inference.
• Environment Variable Parsing: Implemented parsing of required and optional environment variables (mmproj, image, enable_log, ctx_size, n_gpu_layers) to configure the model's execution options.
• Interactive Prompt Handling: Added interactive user input handling and management, including prompt formatting and context resetting mechanisms for continuous dialogue simulation.

wasmedge-ggml/qwen2vl/wasmedge-ggml-qwen2vl.wasm

Potential issues

N/A

Summary of changes

N/A