@sjcrh/proteinpaint-rust 2.148.1 → 2.150.0

package/src/sjprovider.rs

@@ -73,6 +73,15 @@ impl Client {
     pub fn builder() -> ClientBuilder<'static> {
         ClientBuilder::new()
     }
+
+    pub fn completion_model(&self, model: &str) -> CompletionModel {
+        CompletionModel::new(self.clone(), model)
+    }
+
+    pub fn embedding_model(&self, model: &str) -> EmbeddingModel {
+        EmbeddingModel::new(self.clone(), model, 0, self.base_url.to_string())
+    }
+
     pub fn new() -> Self {
         Self::builder().build().expect("Myprovider client should build")
     }
@@ -129,11 +138,16 @@ impl EmbeddingsClient for Client {
 
 impl VerifyClient for Client {
     async fn verify(&self) -> Result<(), VerifyError> {
-        let response = self.get("api/tags").expect("Failed to build request").send().await?;
+        let response = self
+            .get("api/tags")
+            .expect("Failed to build request")
+            .send()
+            .await
+            .unwrap();
         match response.status() {
             reqwest::StatusCode::OK => Ok(()),
             _ => {
-                response.error_for_status()?;
+                response.error_for_status().unwrap();
                 Ok(())
             }
         }
@@ -262,7 +276,7 @@ impl embeddings::EmbeddingModel for EmbeddingModel {
 
             if response.status().is_success() {
                 //println!("response.json:{:?}", response.text().await?);
-                let json_data: Value = serde_json::from_str(&response.text().await?)?;
+                let json_data: Value = serde_json::from_str(&response.text().await.unwrap())?;
                 let emb = json_data["outputs"].as_array().unwrap();
                 //.unwrap_or(&vec![serde_json::Value::String(
                 //    "No embeddings found in json output".to_string(),
@@ -481,12 +495,39 @@ impl CompletionModel {
             panic!("max_new_tokens and top_p not found!");
         };
 
+        let mut user_query = "";
+        let mut system_prompt = "";
+        for message in &full_history {
+            match message {
+                self::Message::User {
+                    content: text,
+                    images: _,
+                    name: _,
+                } => {
+                    //println!("User:{:?}", text);
+                    user_query = text;
+                }
+                self::Message::System {
+                    content: text,
+                    images: _,
+                    name: _,
+                } => {
+                    system_prompt = text;
+                    //println!("System:{:?}", text);
+                }
+                self::Message::Assistant { content: _, id: _ } => {}
+                self::Message::ToolResult { content: _, name: _ } => {}
+            }
+        }
+        let final_text = system_prompt.replace(&"{question}", &user_query);
+
+        //println!("final_text:{:?}", final_text);
         let mut request_payload = json!({
          "inputs":[
                 {
                     "model_name": self.model,
                     "inputs": {
-                        "text": full_history,
+                        "text": final_text,
                         "max_new_tokens": max_new_tokens,
                         "temperature": completion_request.temperature,
                         "top_p": top_p
@@ -612,7 +653,7 @@ impl completion::CompletionModel for CompletionModel {
                 let chunk = match chunk_result {
                     Ok(c) => c,
                     Err(e) => {
-                        yield Err(CompletionError::from(e));
+                        yield Err(CompletionError::RequestError(e.into()));
                         break;
                     }
                 };
@@ -797,7 +838,7 @@ impl ConvertMessage for Message {
                                         images.push(data)
                                     }
                                     rig::message::UserContent::Document(rig::message::Document { data, .. }) => {
-                                        texts.push(data)
+                                        texts.push(data.to_string())
                                     }
                                     _ => {} // Audio not supported by Ollama
                                 }
@@ -993,7 +1034,7 @@ mod tests {
     #[tokio::test]
     #[ignore]
 
-    async fn test_myprovider_implementation() {
+    async fn test_sjprovider_implementation() {
         let user_input = "Generate DE plot for men with weight greater than 30lbs vs women less than 20lbs";
         let serverconfig_file_path = Path::new("../../serverconfig.json");
         let absolute_path = serverconfig_file_path.canonicalize().unwrap();
@@ -1118,17 +1159,17 @@ If a query does not match any of the fields described above, then return JSON wi
         });
 
         // Create RAG agent
-        let agent = AgentBuilder::new(comp_model).preamble("Generate classification for the user query into summary, dge, hierarchial, snv_indel, cnv, variant_calling, sv_fusion and none categories. Return output in JSON with ALWAYS a single word answer { \"answer\": \"dge\" }, that is 'summary' for summary plot, 'dge' for differential gene expression, 'hierarchial' for hierarchial clustering, 'snv_indel' for SNV/Indel, 'cnv' for CNV and 'sv_fusion' for SV/fusion, 'variant_calling' for variant calling, 'surivial' for survival data, 'none' for none of the previously described categories. The answer should always be in lower case").dynamic_context(top_k, vector_store.index(embedding_model)).additional_params(additional).temperature(temperature).build();
+        let agent = AgentBuilder::new(comp_model).preamble("Generate classification for the user query into summary, dge, hierarchial, snv_indel, cnv, variant_calling, sv_fusion and none categories. Return output in JSON with ALWAYS a single word answer { \"answer\": \"dge\" }, that is 'summary' for summary plot, 'dge' for differential gene expression, 'hierarchial' for hierarchial clustering, 'snv_indel' for SNV/Indel, 'cnv' for CNV and 'sv_fusion' for SV/fusion, 'variant_calling' for variant calling, 'surivial' for survival data, 'none' for none of the previously described categories. The answer should always be in lower case. \nQuestion= {question} \nanswer").dynamic_context(top_k, vector_store.index(embedding_model)).additional_params(additional).temperature(temperature).build();
 
         let response = agent.prompt(user_input).await.expect("Failed to prompt myprovider");
 
         //println!("Myprovider: {}", response);
         let result = response.replace("json", "").replace("```", "");
         //println!("result:{}", result);
-        let json_value: Value = serde_json::from_str(&result).expect("REASON");
-        let json_value2: Value = serde_json::from_str(&json_value[0]["generated_text"].to_string()).expect("REASON2");
+        let json_value: Value = serde_json::from_str(&result).expect("REASON2");
+        let json_value2: Value = serde_json::from_str(&json_value[0]["generated_text"].to_string()).expect("REASON3");
         //println!("json_value2:{}", json_value2.as_str().unwrap());
-        let json_value3: Value = serde_json::from_str(&json_value2.as_str().unwrap()).expect("REASON2");
+        let json_value3: Value = serde_json::from_str(&json_value2.as_str().unwrap()).expect("REASON4");
         assert_eq!(json_value3["answer"].to_string().replace("\"", ""), "dge");
     }
 }

package/src/test_ai.rs

@@ -0,0 +1,168 @@
+// For capturing output from a test, run: cd .. && cargo test -- --nocapture
+// Ignored tests: cd .. && export RUST_BACKTRACE=full && time cargo test -- --ignored --nocapture
+#[allow(dead_code)]
+fn main() {}
+
+#[cfg(test)]
+mod tests {
+    use serde_json;
+    use std::fs::{self};
+    use std::path::Path;
+
+    #[derive(PartialEq, Debug, Clone, schemars::JsonSchema, serde::Serialize, serde::Deserialize)]
+    struct ServerConfig {
+        tpmasterdir: String,
+        llm_backend: String,
+        sj_apilink: String,
+        sj_comp_model_name: String,
+        sj_embedding_model_name: String,
+        ollama_apilink: String,
+        ollama_comp_model_name: String,
+        ollama_embedding_model_name: String,
+        genomes: Vec<Genomes>,
+    }
+
+    #[derive(PartialEq, Debug, Clone, schemars::JsonSchema, serde::Serialize, serde::Deserialize)]
+    struct Genomes {
+        name: String,
+        datasets: Vec<Dataset>,
+    }
+
+    #[derive(PartialEq, Debug, Clone, schemars::JsonSchema, serde::Serialize, serde::Deserialize)]
+    struct Dataset {
+        name: String,
+        aifiles: Option<String>, // For now aifiles are defined only for certain datasets
+    }
+
+    #[tokio::test]
+    #[ignore]
+    async fn user_prompts() {
+        let temperature: f64 = 0.01;
+        let max_new_tokens: usize = 512;
+        let top_p: f32 = 0.95;
+        let serverconfig_file_path = Path::new("../../serverconfig.json");
+        let absolute_path = serverconfig_file_path.canonicalize().unwrap();
+
+        // Read the file
+        let data = fs::read_to_string(absolute_path).unwrap();
+
+        // Parse the JSON data
+        let serverconfig: ServerConfig = serde_json::from_str(&data).expect("JSON not in serverconfig.json format");
+
+        for genome in &serverconfig.genomes {
+            for dataset in &genome.datasets {
+                match &dataset.aifiles {
+                    Some(ai_json_file) => {
+                        println!("Testing dataset:{}", dataset.name);
+                        let ai_json_file_path = String::from("../../") + ai_json_file;
+                        let ai_json_file = Path::new(&ai_json_file_path);
+
+                        // Read the file
+                        let ai_data = fs::read_to_string(ai_json_file).unwrap();
+                        // Parse the JSON data
+                        let ai_json: super::super::AiJsonFormat =
+                            serde_json::from_str(&ai_data).expect("AI JSON file does not have the correct format");
+                        //println!("ai_json:{:?}", ai_json);
+                        let genedb = String::from(&serverconfig.tpmasterdir) + &"/" + &ai_json.genedb;
+                        let dataset_db = String::from(&serverconfig.tpmasterdir) + &"/" + &ai_json.db;
+                        let llm_backend_name = &serverconfig.llm_backend;
+                        let llm_backend_type: super::super::llm_backend;
+
+                        if llm_backend_name != "ollama" && llm_backend_name != "SJ" {
+                            panic!(
+                                "This code currently supports only Ollama and SJ provider. llm_backend_name must be \"ollama\" or \"SJ\""
+                            );
+                        } else if *llm_backend_name == "ollama".to_string() {
+                            let ollama_host = &serverconfig.ollama_apilink;
+                            let ollama_embedding_model_name = &serverconfig.ollama_embedding_model_name;
+                            let ollama_comp_model_name = &serverconfig.ollama_comp_model_name;
+                            llm_backend_type = super::super::llm_backend::Ollama();
+                            let ollama_client = super::super::ollama::Client::builder()
+                                .base_url(ollama_host)
+                                .build()
+                                .expect("Ollama server not found");
+                            let embedding_model = ollama_client.embedding_model(ollama_embedding_model_name);
+                            let comp_model = ollama_client.completion_model(ollama_comp_model_name);
+
+                            for chart in ai_json.charts.clone() {
+                                match chart {
+                                    super::super::Charts::Summary(testdata) => {
+                                        for ques_ans in testdata.TestData {
+                                            let user_input = ques_ans.question;
+                                            let llm_output = super::super::run_pipeline(
+                                                &user_input,
+                                                comp_model.clone(),
+                                                embedding_model.clone(),
+                                                llm_backend_type.clone(),
+                                                temperature,
+                                                max_new_tokens,
+                                                top_p,
+                                                &dataset_db,
+                                                &genedb,
+                                                &ai_json,
+                                            )
+                                            .await;
+                                            let mut llm_json_value: super::super::SummaryType = serde_json::from_str(&llm_output.unwrap()).expect("Did not get a valid JSON of type {action: summary, summaryterms:[{clinical: term1}, {geneExpression: gene}], filter:[{term: term1, value: value1}]} from the LLM");
+                                            let mut expected_json_value: super::super::SummaryType = serde_json::from_str(&ques_ans.answer).expect("Did not get a valid JSON of type {action: summary, summaryterms:[{clinical: term1}, {geneExpression: gene}], filter:[{term: term1, value: value1}]} from the LLM");
+                                            assert_eq!(
+                                                llm_json_value.sort_summarytype_struct(),
+                                                expected_json_value.sort_summarytype_struct()
+                                            );
+                                        }
+                                    }
+                                    super::super::Charts::DE(_testdata) => {} // To do
+                                }
+                            }
+                        } else if *llm_backend_name == "SJ".to_string() {
+                            let sjprovider_host = &serverconfig.sj_apilink;
+                            let sj_embedding_model_name = &serverconfig.sj_embedding_model_name;
+                            let sj_comp_model_name = &serverconfig.sj_comp_model_name;
+                            llm_backend_type = super::super::llm_backend::Sj();
+                            let sj_client = super::super::sjprovider::Client::builder()
+                                .base_url(sjprovider_host)
+                                .build()
+                                .expect("SJ server not found");
+                            let embedding_model = sj_client.embedding_model(sj_embedding_model_name);
+                            let comp_model = sj_client.completion_model(sj_comp_model_name);
+
+                            for chart in ai_json.charts.clone() {
+                                match chart {
+                                    super::super::Charts::Summary(testdata) => {
+                                        for ques_ans in testdata.TestData {
+                                            let user_input = ques_ans.question;
+                                            if user_input.len() > 0 {
+                                                let llm_output = super::super::run_pipeline(
+                                                    &user_input,
+                                                    comp_model.clone(),
+                                                    embedding_model.clone(),
+                                                    llm_backend_type.clone(),
+                                                    temperature,
+                                                    max_new_tokens,
+                                                    top_p,
+                                                    &dataset_db,
+                                                    &genedb,
+                                                    &ai_json,
+                                                )
+                                                .await;
+                                                let mut llm_json_value: super::super::SummaryType = serde_json::from_str(&llm_output.unwrap()).expect("Did not get a valid JSON of type {action: summary, summaryterms:[{clinical: term1}, {geneExpression: gene}], filter:[{term: term1, value: value1}]} from the LLM");
+                                                let mut expected_json_value: super::super::SummaryType = serde_json::from_str(&ques_ans.answer).expect("Did not get a valid JSON of type {action: summary, summaryterms:[{clinical: term1}, {geneExpression: gene}], filter:[{term: term1, value: value1}]} from the LLM");
+                                                assert_eq!(
+                                                    llm_json_value.sort_summarytype_struct(),
+                                                    expected_json_value.sort_summarytype_struct()
+                                                );
+                                            } else {
+                                                panic!("The user input is empty");
+                                            }
+                                        }
+                                    }
+                                    super::super::Charts::DE(_testdata) => {} // To do
+                                }
+                            }
+                        }
+                    }
+                    None => {}
+                }
+            }
+        }
+    }
+}