changed summary logic

Former-commit-id: d509673b3e597bee08e193e8208ae263c3e35a33

'

@@ -0,0 +1 @@
+Unable to initialize device PRN

0.41.0

@@ -0,0 +1,19 @@
+Collecting bitsandbytes
+  Downloading bitsandbytes-0.47.0-py3-none-win_amd64.whl.metadata (11 kB)
+Requirement already satisfied: torch<3,>=2.2 in c:\users\sachin\appdata\local\programs\python\python311\lib\site-packages (from bitsandbytes) (2.3.0)
+Requirement already satisfied: numpy>=1.17 in c:\users\sachin\appdata\local\programs\python\python311\lib\site-packages (from bitsandbytes) (1.24.3)
+Requirement already satisfied: filelock in c:\users\sachin\appdata\local\programs\python\python311\lib\site-packages (from torch<3,>=2.2->bitsandbytes) (3.18.0)
+Requirement already satisfied: typing-extensions>=4.8.0 in c:\users\sachin\appdata\local\programs\python\python311\lib\site-packages (from torch<3,>=2.2->bitsandbytes) (4.14.1)
+Requirement already satisfied: sympy in c:\users\sachin\appdata\local\programs\python\python311\lib\site-packages (from torch<3,>=2.2->bitsandbytes) (1.14.0)
+Requirement already satisfied: networkx in c:\users\sachin\appdata\local\programs\python\python311\lib\site-packages (from torch<3,>=2.2->bitsandbytes) (3.5)
+Requirement already satisfied: jinja2 in c:\users\sachin\appdata\local\programs\python\python311\lib\site-packages (from torch<3,>=2.2->bitsandbytes) (3.1.6)
+Requirement already satisfied: fsspec in c:\users\sachin\appdata\local\programs\python\python311\lib\site-packages (from torch<3,>=2.2->bitsandbytes) (2025.7.0)
+Requirement already satisfied: mkl<=2021.4.0,>=2021.1.1 in c:\users\sachin\appdata\local\programs\python\python311\lib\site-packages (from torch<3,>=2.2->bitsandbytes) (2021.4.0)
+Requirement already satisfied: intel-openmp==2021.* in c:\users\sachin\appdata\local\programs\python\python311\lib\site-packages (from mkl<=2021.4.0,>=2021.1.1->torch<3,>=2.2->bitsandbytes) (2021.4.0)
+Requirement already satisfied: tbb==2021.* in c:\users\sachin\appdata\local\programs\python\python311\lib\site-packages (from mkl<=2021.4.0,>=2021.1.1->torch<3,>=2.2->bitsandbytes) (2021.13.1)
+Requirement already satisfied: MarkupSafe>=2.0 in c:\users\sachin\appdata\local\programs\python\python311\lib\site-packages (from jinja2->torch<3,>=2.2->bitsandbytes) (2.1.5)
+Requirement already satisfied: mpmath<1.4,>=1.1.0 in c:\users\sachin\appdata\local\programs\python\python311\lib\site-packages (from sympy->torch<3,>=2.2->bitsandbytes) (1.3.0)
+Downloading bitsandbytes-0.47.0-py3-none-win_amd64.whl (60.7 MB)
+   ---------------------------------------- 60.7/60.7 MB 12.1 MB/s  0:00:05
+Installing collected packages: bitsandbytes
+Successfully installed bitsandbytes-0.47.0

ai_med_extract/api/__pycache__/routes.cpython-311.pyc.REMOVED.git-id

@@ -0,0 +1 @@
+2129106af05cccbba389f4e2c6915fad2060edfd

ai_med_extract/api/routes.py

@@ -260,6 +260,79 @@ def get_summarizer_pipeline(summarizer_model_type, summarizer_model_name):
 
 
 def register_routes(app, agents):
+    from ai_med_extract.utils.openvino_summarizer_utils import (
+        parse_ehr_chartsummarydtl, visits_sorted, compute_deltas, build_compact_baseline, delta_to_text, build_main_prompt, validate_and_compare_summaries
+    )
+    import threading
+    patient_summary_state = {}
+    state_lock = threading.Lock()
+
+    @app.route('/api/patient_summary_openvino', methods=['POST'])
+    def patient_summary_openvino():
+        """
+        Generate a patient summary using OpenVINO-style prompt, delta, and validation logic.
+        Accepts EHR API response JSON (or just chartsummarydtl) and returns summary and validation.
+        """
+        try:
+            data = request.get_json()
+            ehr_result = data.get("result") or data
+            chartsummarydtl = ehr_result.get("chartsummarydtl") if isinstance(ehr_result, dict) else None
+            if not chartsummarydtl:
+                return jsonify({"error": "Missing chartsummarydtl in input"}), 400
+
+            # Normalize visits
+            visits = parse_ehr_chartsummarydtl(chartsummarydtl)
+            patient_id = ehr_result.get("patientid") or ehr_result.get("patientnumber") or "default"
+
+            # Track state per patient (in-memory, thread-safe)
+            with state_lock:
+                state = patient_summary_state.setdefault(patient_id, {"visits": [], "last_summary": ""})
+                old_visits = state["visits"]
+                old_summary = state["last_summary"]
+
+            # Compute deltas and prompt
+            delta = compute_deltas(old_visits, visits)
+            all_visits = visits_sorted(old_visits + visits)
+            baseline = build_compact_baseline(all_visits)
+            delta_text = delta_to_text(delta)
+            prompt = build_main_prompt(old_summary, baseline, delta_text)
+
+            # Model selection logic (model_name, model_type)
+            model_name = data.get("model_name") or "microsoft/Phi-3-mini-4k-instruct"
+            model_type = data.get("model_type") or "text-generation"
+            # Use existing model loader abstraction
+            if model_type == "text-generation":
+                loader = agents.get("medical_data_extractor")
+            else:
+                loader = agents.get("summarizer")
+            pipeline = loader.model_loader.load() if hasattr(loader, "model_loader") else None
+            if not pipeline:
+                return jsonify({"error": "Model pipeline not available"}), 500
+
+            # Run inference
+            import torch
+            torch.set_num_threads(2)
+            inputs = pipeline.tokenizer([prompt], return_tensors="pt")
+            outputs = pipeline.model.generate(**inputs, max_new_tokens=400, do_sample=False, pad_token_id=pipeline.tokenizer.eos_token_id or 32000)
+            text = pipeline.tokenizer.decode(outputs[0], skip_special_tokens=True)
+            new_summary = text.split("Now generate the complete, updated clinical summary with all four sections:")[-1].strip()
+
+            # Update state
+            with state_lock:
+                state["visits"] = all_visits
+                state["last_summary"] = new_summary
+
+            # Validation
+            validation_report = validate_and_compare_summaries(old_summary, new_summary, "Update")
+
+            return jsonify({
+                "summary": new_summary,
+                "validation": validation_report,
+                "baseline": baseline,
+                "delta": delta_text
+            }), 200
+        except Exception as e:
+            return jsonify({"error": f"Failed to generate summary: {str(e)}"}), 500
     # Configure upload directory based on environment
     import os
 
@@ -902,16 +975,25 @@ def register_routes(app, agents):
     
     @app.route('/generate_patient_summary', methods=['POST'])
     def generate_patient_summary():
+        """
+        Enhanced: Uses OpenVINO-style prompt, delta, and validation logic for patient summary generation.
+        """
+        from ai_med_extract.utils.openvino_summarizer_utils import (
+            parse_ehr_chartsummarydtl, visits_sorted, compute_deltas, build_compact_baseline, delta_to_text, build_main_prompt, validate_and_compare_summaries
+        )
+        import threading
+        if not hasattr(generate_patient_summary, "state"):
+            generate_patient_summary.state = {}
+            generate_patient_summary.lock = threading.Lock()
+        state = generate_patient_summary.state
+        state_lock = generate_patient_summary.lock
         try:
             data = request.get_json()
             patientid = data.get("patientid")
             token = data.get("token")
             key = data.get("key")
-            # Be robust to null/empty model names provided by clients
-            model_name = data.get("patient_summarizer_model_name")
-            if not model_name or str(model_name).strip().lower() in {"", "none", "null"}:
-                model_name = "falconsai/medical_summarization"
-
+            model_name = data.get("patient_summarizer_model_name") or "falconsai/medical_summarization"
+            model_type = data.get("patient_summarizer_model_type") or data.get("model_type") or "summarization"
             if not patientid or not token or not key:
                 return jsonify({"error": "Missing required fields: patientid, token, or key"}), 400
 
@@ -920,9 +1002,8 @@ def register_routes(app, agents):
                 "Authorization": f"Bearer {token}",
                 "x-api-key": key,
                 "Content-Type": "application/json",
-                }
+            }
             body = json.dumps({"patientid": patientid})
-
             response = requests.post(api_url, data=body, headers=headers, timeout=30)
             if response.status_code != 200:
                 return jsonify({
@@ -930,77 +1011,71 @@ def register_routes(app, agents):
                     "status": response.status_code,
                     "message": response.text
                 }), 502
-
-            # Be tolerant to any response format: JSON, text, key-values, etc.
             try:
                 api_data = response.json()
             except ValueError:
                 api_data = response.text
-
-            # If dict, prefer nested 'result'; otherwise pass through
             if isinstance(api_data, dict):
-                raw_data = api_data.get("result") or api_data
+                ehr_result = api_data.get("result") or api_data
             else:
-                raw_data = api_data
-
-            logger.info(f"Fetched data for patient {patientid}")
-            cleaned_data = clean_patient_data(raw_data)
-            if not cleaned_data or "result" not in cleaned_data:
-                return jsonify({"error": "Failed to clean patient data"}), 500
-
-            # Helper: convert plain text sections to Markdown
-            def _to_markdown_summary(text: str) -> str:
-                headings = [
-                    "Patient Overview",
-                    "Visit History",
-                    "Trend Analysis",
-                    "Assessment",
-                    "Recommendations",
-                ]
-                lines_out = []
-                for ln in str(text or "").splitlines():
-                    s = ln.strip()
-                    if not s:
-                        lines_out.append("")
-                        continue
-                    matched = next((h for h in headings if s.lower().startswith(h.lower() + ":")), None)
-                    if matched:
-                        after = s[len(matched) + 1:].strip()
-                        lines_out.append(f"### {matched}")
-                        if after:
-                            lines_out.append(after)
-                        continue
-                    if s.startswith("- "):
-                        lines_out.append(s)
-                    elif ":" in s and not s.startswith("###"):
-                        lines_out.append(f"- {s}")
-                    else:
-                        lines_out.append(s)
-                return "\n".join(lines_out).strip()
-
-            # Pass full cleaned_data (with timeline) to agent
-            summarizer = PatientSummarizerAgent(model_name=model_name)
-            raw_summary = summarizer.generate_clinical_summary(cleaned_data)
-            summary_md = _to_markdown_summary(raw_summary)
-
-            if torch.cuda.is_available():
-                torch.cuda.empty_cache()
-
-            # Optional: return flattened for debugging
-            flattened = flatten_to_string_list(cleaned_data)
-
+                ehr_result = api_data
+            chartsummarydtl = ehr_result.get("chartsummarydtl") if isinstance(ehr_result, dict) else None
+            if not chartsummarydtl:
+                return jsonify({"error": "Missing chartsummarydtl in EHR response"}), 500
+            visits = parse_ehr_chartsummarydtl(chartsummarydtl)
+            # Per-patient state (in-memory)
+            with state_lock:
+                patient_state = state.setdefault(patientid, {"visits": [], "last_summary": ""})
+                old_visits = patient_state["visits"]
+                old_summary = patient_state["last_summary"]
+            delta = compute_deltas(old_visits, visits)
+            all_visits = visits_sorted(old_visits + visits)
+            baseline = build_compact_baseline(all_visits)
+            delta_text = delta_to_text(delta)
+            prompt = build_main_prompt(old_summary, baseline, delta_text)
+            # Model selection logic (supporting OpenVINO and HuggingFace)
+            pipeline = None
+            loader = None
+            import torch
+            torch.set_num_threads(2)
+            if model_type in {"text-generation", "causal-openvino"}:
+                # Try to use an existing loader if available
+                loader = agents.get("medical_data_extractor")
+                if not loader or getattr(loader, 'model_name', None) != model_name:
+                    # Dynamically create OpenVINO loader if needed
+                    from ai_med_extract.utils.model_loader_spaces import get_openvino_pipeline
+                    try:
+                        pipeline = get_openvino_pipeline(model_name)
+                    except Exception as e:
+                        return jsonify({"error": f"Failed to load OpenVINO pipeline: {str(e)}"}), 500
+            elif model_type == "summarization":
+                loader = agents.get("summarizer")
+            # Use loader if available
+            if not pipeline and loader and hasattr(loader, "model_loader"):
+                pipeline = loader.model_loader.load()
+            if not pipeline:
+                return jsonify({"error": "Model pipeline not available"}), 500
+            inputs = pipeline.tokenizer([prompt], return_tensors="pt")
+            outputs = pipeline.model.generate(**inputs, max_new_tokens=400, do_sample=False, pad_token_id=pipeline.tokenizer.eos_token_id or 32000)
+            text = pipeline.tokenizer.decode(outputs[0], skip_special_tokens=True)
+            new_summary = text.split("Now generate the complete, updated clinical summary with all four sections:")[-1].strip()
+            with state_lock:
+                patient_state["visits"] = all_visits
+                patient_state["last_summary"] = new_summary
+            validation_report = validate_and_compare_summaries(old_summary, new_summary, "Update")
             return jsonify({
-                "summary": summary_md,
-                "flattened": flattened
+                "summary": new_summary,
+                "validation": validation_report,
+                "baseline": baseline,
+                "delta": delta_text
             }), 200
-
         except requests.exceptions.Timeout:
             return jsonify({"error": "Request to EHR API timed out"}), 504
         except requests.exceptions.RequestException as e:
             return jsonify({"error": f"Network error: {str(e)}"}), 503
         except Exception as e:
             logger.error(f"Unexpected error: {str(e)}", exc_info=True)
-            return jsonify({"error": f"Internal server error: {str(e)}"}), 500  
+            return jsonify({"error": f"Internal server error: {str(e)}"}), 500
         
     @app.route("/")
     def home():

ai_med_extract/utils/model_loader_spaces.py

@@ -0,0 +1,22 @@
+import torch
+from transformers import AutoTokenizer
+from optimum.intel.openvino import OVModelForCausalLM
+
+class OpenVinoPipeline:
+	def __init__(self, model, tokenizer):
+		self.model = model
+		self.tokenizer = tokenizer
+
+def get_openvino_pipeline(model_name: str):
+	"""
+	Loads an OpenVINO CausalLM pipeline for the given model name or IR directory.
+	"""
+	# If model_name is a directory, try to load IR from there; else, download and export
+	import os
+	if os.path.isdir(model_name):
+		model = OVModelForCausalLM.from_pretrained(model_name, compile=True, device="CPU")
+		tokenizer = AutoTokenizer.from_pretrained(model_name, use_fast=True)
+	else:
+		model = OVModelForCausalLM.from_pretrained(model_name, export=True, compile=True, device="CPU")
+		tokenizer = AutoTokenizer.from_pretrained(model_name, use_fast=True)
+	return OpenVinoPipeline(model, tokenizer)

ai_med_extract/utils/openvino_summarizer_utils.py

@@ -0,0 +1,162 @@
+import os
+import json
+import re
+import textwrap
+import difflib
+import logging
+from copy import deepcopy
+
+def parse_ehr_chartsummarydtl(chartsummarydtl):
+    """
+    Converts EHR API chartsummarydtl list to the internal visit format expected by the summarizer.
+    """
+    visits = []
+    for entry in chartsummarydtl:
+        visit = {}
+        # Parse chartdate
+        visit["chartdate"] = entry.get("chartdate", "")[:10]  # YYYY-MM-DD
+        # Parse vitals
+        vitals_dict = {}
+        weight = None
+        if "vitals" in entry:
+            for v in entry["vitals"]:
+                if ":" in v:
+                    k, val = v.split(":", 1)
+                    k = k.strip()
+                    val = val.strip()
+                    if k.lower().startswith("weight"):
+                        weight = val
+                    else:
+                        vitals_dict[k] = val
+        visit["vitals"] = vitals_dict
+        if weight:
+            visit["weight"] = weight
+        # Allergies
+        if "allergies" in entry:
+            visit["allergies"] = entry["allergies"]
+        # Diagnosis
+        if "diagnosis" in entry:
+            visit["diagnosis"] = entry["diagnosis"]
+        # Medications
+        if "medications" in entry:
+            visit["medications"] = entry["medications"]
+        # Labtests
+        labtests = []
+        if "labtests" in entry:
+            for l in entry["labtests"]:
+                name = l.get("name", "")
+                value = l.get("value", "")
+                if name or value:
+                    labtests.append({"name": name, "value": value})
+        visit["labtests"] = labtests
+        # Radiology orders
+        if "radiologyorders" in entry:
+            visit["radiologyorders"] = [r.get("name", "") for r in entry["radiologyorders"] if r.get("name")]
+        visits.append(visit)
+    return visits
+
+# ========== PROMPT, DELTA, VALIDATION LOGIC (adapted from your script) =============
+ALIASES = {("vitals","Bp(sys)(mmHg)"): [("vitals","Bp(sys)(mmHg)"), ("vitals","Bp_sys"), ("vitals","SBP")],
+           ("vitals","Bp(dia)(mmHg)"): [("vitals","Bp(dia)(mmHg)"), ("vitals","Bp_dia"), ("vitals","DBP")],
+           ("labtests","HbA1c (%)"): [("labtests","HbA1c (%)"), ("labtests","HbA1c")],
+           ("labtests","Creatinine Ratio"): [("labtests","Creatinine Ratio"), ("labtests","Creatinine")],
+}
+def visits_sorted(v):
+    return sorted(v, key=lambda v: v.get("chartdate", ""))
+def to_float(val):
+    try:
+        s = str(val); m = re.findall(r"-?\d+\.?\d*", s)
+        return float(m[0]) if m else None
+    except: return None
+def _latest_value_exact(visits, key_path):
+    v_sorted = visits_sorted(visits)
+    if not v_sorted: return None
+    if key_path[0] == "labtests":
+        for v in reversed(v_sorted):
+            for lab in v.get("labtests", []):
+                if lab.get("name") == key_path[1]: return lab.get("value")
+        return None
+    for v in reversed(v_sorted):
+        cur = v; ok = True
+        for k in key_path:
+            if isinstance(cur, dict) and k in cur: cur = cur[k]
+            else: ok=False; break
+        if ok: return cur
+    return None
+def latest_value(visits, key_path):
+    for kp in ALIASES.get(key_path, [key_path]):
+        val = _latest_value_exact(visits, kp)
+        if val is not None: return val
+    return None
+def active_set(visits, field):
+    s = set()
+    for v in visits: s.update(v.get(field, []))
+    return s
+def _fmt(x, spec=None):
+    if x is None: return "N/A"
+    try: return format(x, spec) if spec else str(x)
+    except Exception: return str(x)
+def compute_deltas(old_visits, new_visits):
+    prev_all = old_visits
+    curr_all = old_visits + new_visits
+    def get_val(visits, path): return to_float(latest_value(visits, path))
+    w_p, w_c = get_val(prev_all, ("weight",)), get_val(curr_all, ("weight",))
+    s_p, s_c = get_val(prev_all, ("vitals","Bp(sys)(mmHg)")), get_val(curr_all, ("vitals","Bp(sys)(mmHg)"))
+    d_p, d_c = get_val(prev_all, ("vitals","Bp(dia)(mmHg)")), get_val(curr_all, ("vitals","Bp(dia)(mmHg)"))
+    h_p, h_c = get_val(prev_all, ("labtests","HbA1c (%)")), get_val(curr_all, ("labtests","HbA1c (%)"))
+    c_p, c_c = get_val(prev_all, ("labtests","Creatinine Ratio")), get_val(curr_all, ("labtests","Creatinine Ratio"))
+    return {
+        "added_dx": sorted(list(active_set(curr_all,"diagnosis") - active_set(prev_all,"diagnosis"))),
+        "started_meds": sorted(list(active_set(curr_all,"medications") - active_set(prev_all,"medications"))),
+        "stopped_meds": sorted(list(active_set(prev_all,"medications") - active_set(curr_all,"medications"))),
+        "weight": {"prev": w_p, "curr": w_c, "delta": (w_c - w_p) if w_p and w_c else None},
+        "bp_sys": {"prev": s_p, "curr": s_c, "delta": (s_c - s_p) if s_p and s_c else None},
+        "bp_dia": {"prev": d_p, "curr": d_c, "delta": (d_c - d_p) if d_p and d_c else None},
+        "hba1c": {"prev": h_p, "curr": h_c, "delta": (h_c - h_p) if h_p and h_c else None},
+        "cratio": {"prev": c_p, "curr": c_c, "delta": (c_c - c_p) if c_p and c_c else None},
+    }
+def build_compact_baseline(all_visits):
+    return f"Latest date: {latest_value(all_visits,('chartdate',)) or 'N/A'}\n" \
+           f"Active Diagnoses: {', '.join(sorted(active_set(all_visits,'diagnosis'))) or 'N/A'}\n" \
+           f"Active Medications: {', '.join(sorted(active_set(all_visits,'medications'))) or 'N/A'}\n" \
+           f"Latest Vitals: Bp: {latest_value(all_visits,('vitals','Bp(sys)(mmHg)'))}/{latest_value(all_visits,('vitals','Bp(dia)(mmHg)'))} mmHg, Weight: {latest_value(all_visits,('weight',))}\n" \
+           f"Latest Labs: HbA1c: {latest_value(all_visits,('labtests','HbA1c (%)'))}%, Creatinine: {latest_value(all_visits,('labtests','Creatinine Ratio'))}"
+def delta_to_text(delta):
+    L = []
+    if delta["added_dx"]: L.append("New Diagnoses: " + ", ".join(delta["added_dx"]))
+    if delta["started_meds"]: L.append("Medications Started: " + ", ".join(delta["started_meds"]))
+    if delta["stopped_meds"]: L.append("Medications Stopped: " + ", ".join(delta["stopped_meds"]))
+    w = delta["weight"]; L.append(f"Weight: {_fmt(w['prev'])} -> {_fmt(w['curr'])} (Δ {_fmt(w['delta'], '+.1f')})")
+    s, d = delta["bp_sys"], delta["bp_dia"]; L.append(f"BP: {_fmt(s['curr'])}/{_fmt(d['curr'])} (Δs {_fmt(s['delta'], '+.0f')}, Δd {_fmt(d['delta'], '+.0f')})")
+    h, c = delta["hba1c"], delta["cratio"]; L.append(f"HbA1c: {_fmt(h['prev'])} -> {_fmt(h['curr'])} (Δ {_fmt(h['delta'], '.1f')})"); L.append(f"Creatinine: {_fmt(c['prev'])} -> {_fmt(c['curr'])} (Δ {_fmt(c['delta'], '.1f')})")
+    return "\n".join(L)
+def build_main_prompt(prev_summary, baseline, delta_text):
+    prev_excerpt = textwrap.shorten(prev_summary, width=700, placeholder="...") if prev_summary else "None"
+    return (
+        "You are an expert clinical AI assistant. Your task is to update a patient summary.\n"
+        "Use the PRIOR SUMMARY for context. The STRUCTURED BASELINE and DELTAS are the absolute ground truth.\n"
+        "Produce a concise, physician-ready update. Never omit critical new information from the deltas.\n\n"
+        "The summary MUST have four sections:\n"
+        "1) Clinical Assessment\n"
+        "2) Key Trends & Changes\n"
+        "3) Plan & Suggested Actions\n"
+        "4) Direct Guidance for Physician\n\n"
+        f"PRIOR SUMMARY (context):\n{prev_excerpt}\n\n"
+        f"STRUCTURED BASELINE (authoritative):\n{baseline}\n\n"
+        f"STRUCTURED DELTAS (authoritative):\n{delta_text}\n\n"
+        "Now generate the complete, updated clinical summary with all four sections:"
+    )
+def validate_and_compare_summaries(old_summary, new_summary, update_name=""):
+    report = f"### Validation Report for {update_name}\n"
+    report += "This report validates that the updated summary incorporates new information correctly.\n"
+    report += "\n**Unified Diff (Line-by-Line Changes):**\n"
+    diff = difflib.unified_diff(
+        old_summary.splitlines(), new_summary.splitlines(),
+        fromfile='Previous Summary', tofile='Current Summary', lineterm=''
+    )
+    diff_text = "\n".join(list(diff))
+    if not diff_text:
+        report += "No textual differences found between summaries.\n"
+    else:
+        report += "```diff\n" + diff_text + "\n```\n"
+    return report

export_phi3_openvino.py

@@ -0,0 +1,9 @@
+from optimum.intel.openvino import OVModelForCausalLM
+from transformers import AutoTokenizer
+
+model_id = "microsoft/Phi-3-mini-4k-instruct"
+export_dir = "ov_models/microsoft_Phi-3-mini-4k-instruct_ir"
+
+tokenizer = AutoTokenizer.from_pretrained(model_id, library_name="transformers")
+model = OVModelForCausalLM.from_pretrained(model_id, export=True, save_dir=export_dir)
+tokenizer.save_pretrained(export_dir)

old_requirements.txt

@@ -0,0 +1,165 @@
+about-time==4.2.1
+accelerate==0.25.0
+aiofiles==23.2.1
+aiohappyeyeballs==2.6.1
+aiohttp==3.12.15
+aiosignal==1.4.0
+alive-progress==3.3.0
+altair==5.5.0
+annotated-types==0.7.0
+anyio==4.10.0
+attrs==25.3.0
+autograd==1.8.0
+bitsandbytes==0.47.0
+blinker==1.9.0
+blis==0.7.11
+catalogue==2.0.10
+certifi==2025.8.3
+cffi==1.17.1
+charset-normalizer==3.4.3
+click==8.2.1
+cloudpathlib==0.16.0
+cma==4.3.0
+colorama==0.4.6
+confection==0.1.5
+contourpy==1.3.2
+cryptography==45.0.6
+cycler==0.12.1
+cymem==2.0.11
+datasets==4.0.0
+Deprecated==1.2.18
+dill==0.3.8
+einops==0.7.0
+en-core-web-sm @ https://github.com/explosion/spacy-models/releases/download/en_core_web_sm-3.7.1/en_core_web_sm-3.7.1-py3-none-any.whl#sha256=86cc141f63942d4b2c5fcee06630fd6f904788d2f0ab005cce45aadb8fb73889
+fastapi==0.116.1
+ffmpy==0.6.1
+filelock==3.18.0
+Flask==3.1.0
+flask-cors==5.0.1
+fonttools==4.59.0
+frozenlist==1.7.0
+fsspec==2025.3.0
+gradio==4.13.0
+gradio_client==0.8.0
+graphemeu==0.7.2
+gunicorn==21.2.0
+h11==0.16.0
+httpcore==1.0.9
+httpx==0.28.1
+huggingface-hub==0.34.4
+idna==3.10
+importlib_resources==6.5.2
+intel-openmp==2021.4.0
+itsdangerous==2.2.0
+Jinja2==3.1.6
+joblib==1.5.1
+jsonschema==4.25.0
+jsonschema-specifications==2025.4.1
+kiwisolver==1.4.9
+langcodes==3.5.0
+language_data==1.3.0
+llvmlite==0.44.0
+lxml==6.0.0
+marisa-trie==1.2.1
+markdown-it-py==4.0.0
+MarkupSafe==2.1.5
+matplotlib==3.10.5
+mdurl==0.1.2
+mkl==2021.4.0
+more-itertools==10.7.0
+mpmath==1.3.0
+multidict==6.6.4
+multiprocess==0.70.16
+murmurhash==1.0.13
+narwhals==2.1.1
+natsort==8.4.0
+networkx==3.4.2
+ninja==1.11.1.4
+nltk==3.8.1
+nncf==2.17.0
+numba==0.61.2
+numpy==1.24.3
+onnx==1.18.0
+openai-whisper==20231117
+opencv-python-headless==4.8.1.78
+openvino==2025.2.0
+openvino-telemetry==2025.2.0
+openvino-tokenizers==2025.2.0.1
+optimum==1.27.0
+optimum-intel==1.25.2
+orjson==3.11.2
+packaging==25.0
+pandas==2.1.4
+pdf2image==1.16.3
+pdfminer.six==20221105
+pdfplumber==0.10.3
+Pillow==10.1.0
+preshed==3.0.10
+propcache==0.3.2
+protobuf==4.25.1
+psutil==7.0.0
+pyarrow==21.0.0
+pycparser==2.22
+pydantic==2.11.7
+pydantic_core==2.33.2
+pydot==3.0.4
+pydub==0.25.1
+Pygments==2.19.2
+pymoo==0.6.1.5
+pyparsing==3.2.3
+PyPDF2==3.0.1
+pypdfium2==4.30.0
+pytesseract==0.3.10
+python-dateutil==2.9.0.post0
+python-docx==1.0.1
+python-dotenv==1.0.1
+python-multipart==0.0.20
+pytz==2025.2
+PyYAML==6.0.2
+referencing==0.36.2
+regex==2025.7.34
+requests==2.32.5
+rich==13.9.4
+rpds-py==0.27.0
+safetensors==0.6.2
+scikit-learn==1.3.2
+scipy==1.11.4
+semantic-version==2.10.0
+sentence-transformers==5.1.0
+sentencepiece==0.1.99
+shellingham==1.5.4
+six==1.17.0
+smart-open==6.4.0
+sniffio==1.3.1
+spacy==3.7.2
+spacy-legacy==3.0.12
+spacy-loggers==1.0.5
+srsly==2.5.1
+starlette==0.47.2
+sympy==1.14.0
+tabulate==0.9.0
+tbb==2021.13.1
+termcolor==3.1.0
+thinc==8.2.5
+threadpoolctl==3.6.0
+tiktoken==0.5.2
+tokenizers==0.21.4
+tomlkit==0.12.0
+torch==2.3.0
+torchaudio==2.3.0
+torchvision==0.18.0
+tqdm==4.67.1
+transformers==4.53.3
+typer==0.9.4
+typing-inspection==0.4.1
+typing_extensions==4.14.1
+tzdata==2025.2
+urllib3==2.5.0
+uvicorn==0.35.0
+wasabi==1.1.3
+weasel==0.3.4
+websockets==11.0.3
+Werkzeug==3.1.3
+wrapt==1.17.3
+xxhash==3.5.0
+yarl==1.20.1

ov_models/microsoft_Phi-3-mini-4k-instruct_ir/chat_template.jinja

@@ -0,0 +1,8 @@
+{% for message in messages %}{% if message['role'] == 'system' %}{{'<|system|>
+' + message['content'] + '<|end|>
+'}}{% elif message['role'] == 'user' %}{{'<|user|>
+' + message['content'] + '<|end|>
+'}}{% elif message['role'] == 'assistant' %}{{'<|assistant|>
+' + message['content'] + '<|end|>
+'}}{% endif %}{% endfor %}{% if add_generation_prompt %}{{ '<|assistant|>
+' }}{% else %}{{ eos_token }}{% endif %}

ov_models/microsoft_Phi-3-mini-4k-instruct_ir/special_tokens_map.json

@@ -0,0 +1,30 @@
+{
+  "bos_token": {
+    "content": "<s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "unk_token": {
+    "content": "<unk>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

ov_models/microsoft_Phi-3-mini-4k-instruct_ir/tokenizer.json.REMOVED.git-id

@@ -0,0 +1 @@
+759de6dd15d187c9ececdea11d3287d4cb4b604a