Merge pull request #37 from nomyo-ai/dev-v0.7.x-semcache

Dev v0.7.x semcache addtl. feature

requirements.txt

@@ -32,6 +32,7 @@ PyYAML==6.0.3
 sniffio==1.3.1
 starlette==0.49.1
 truststore==0.10.4
+tiktoken==0.12.0
 tqdm==4.67.1
 typing-inspection==0.4.1
 typing_extensions==4.14.1

router.py

@@ -78,6 +78,107 @@ def _mask_secrets(text: str) -> str:
     text = re.sub(r"(?i)(api[-_ ]key\s*[:=]\s*)([^\s]+)", r"\1***redacted***", text)
     return text
 
+# ------------------------------------------------------------------
+# Context-window sliding-window helpers
+# ------------------------------------------------------------------
+try:
+    import tiktoken as _tiktoken
+    _tiktoken_enc = _tiktoken.get_encoding("cl100k_base")
+except Exception:
+    _tiktoken_enc = None
+
+def _count_message_tokens(messages: list) -> int:
+    """Approximate token count for a message list.
+
+    Uses tiktoken cl100k_base when available (within ~5-15% of llama tokenizers).
+    Falls back to char/4 heuristic if tiktoken is unavailable.
+    Formula follows OpenAI's per-message overhead: 4 tokens/message + content + 2 priming.
+    """
+    if _tiktoken_enc is None:
+        return sum(len(str(m.get("content", ""))) for m in messages) // 4
+
+    total = 2  # priming tokens
+    for msg in messages:
+        total += 4  # per-message role/separator overhead
+        content = msg.get("content", "")
+        if isinstance(content, str):
+            total += len(_tiktoken_enc.encode(content))
+        elif isinstance(content, list):
+            for part in content:
+                if isinstance(part, dict) and part.get("type") == "text":
+                    total += len(_tiktoken_enc.encode(part.get("text", "")))
+    return total
+
+def _trim_messages_for_context(
+    messages: list,
+    n_ctx: int,
+    safety_margin: int = None,
+    target_tokens: int = None,
+) -> list:
+    """Sliding-window trim — mirrors what llama.cpp context-shift used to do.
+
+    Keeps all system messages and the most recent non-system messages that fit
+    within (n_ctx - safety_margin) tokens. Oldest non-system messages are dropped
+    first (FIFO). The last message is always preserved.
+
+    safety_margin defaults to 1/4 of n_ctx to leave headroom for the generated
+    response, including RAG tool results and tool call JSON synthesis.
+
+    target_tokens: if provided, overrides the (n_ctx - safety_margin) target.
+    Pass a calibrated value when actual n_prompt_tokens is known from the error
+    body so that tiktoken underestimation vs the backend tokenizer is corrected.
+    """
+    if target_tokens is not None:
+        target = target_tokens
+    else:
+        if safety_margin is None:
+            safety_margin = n_ctx // 4
+        target = n_ctx - safety_margin
+    system_msgs = [m for m in messages if m.get("role") == "system"]
+    non_system = [m for m in messages if m.get("role") != "system"]
+
+    while len(non_system) > 1:
+        if _count_message_tokens(system_msgs + non_system) <= target:
+            break
+        non_system.pop(0)  # drop oldest non-system message
+
+    # Ensure the first non-system message is a user message (chat templates require it).
+    # Drop any leading assistant/tool messages that were left after trimming.
+    while non_system and non_system[0].get("role") != "user":
+        non_system.pop(0)
+
+    return system_msgs + non_system
+
+
+
+def _calibrated_trim_target(msgs: list, n_ctx: int, actual_tokens: int) -> int:
+    """Return a tiktoken-scale trim target based on how much backend tokens must be shed.
+
+    actual_tokens includes messages + tool schemas + overhead as counted by the backend.
+    _count_message_tokens only counts message text, so we cannot derive an accurate
+    per-token scale from the ratio.  Instead we compute the *delta* we need to remove
+    in backend space, then convert just that delta to tiktoken scale (×1.2 buffer).
+
+    Example: actual=17993, n_ctx=16384, headroom=4096 → need to shed 5705 backend
+    tokens → shed 6846 tiktoken tokens from messages.
+    """
+    cur_tiktoken = _count_message_tokens(msgs)
+    headroom = n_ctx // 4  # reserve for generated output
+    max_prompt = n_ctx - headroom  # desired max backend tokens in prompt
+    to_shed = max(0, actual_tokens - max_prompt)  # backend tokens we must drop
+    # Convert to tiktoken scale with 20% buffer (tiktoken underestimates llama by ~15-20%)
+    tiktoken_to_shed = int(to_shed * 1.2)
+    return max(1, cur_tiktoken - tiktoken_to_shed)
+
+# Per-(endpoint, model) n_ctx cache.
+# Populated from two sources:
+#   1. 400 exceed_context_size_error body → n_ctx field
+#   2. finish_reason/done_reason == "length" in streaming → prompt_tokens + completion_tokens
+# Only used for proactive pre-trimming when n_ctx <= _CTX_TRIM_SMALL_LIMIT,
+# so large-context models (200k+ for coding) are never touched.
+_endpoint_nctx: dict[tuple[str, str], int] = {}
+_CTX_TRIM_SMALL_LIMIT = 32768  # only proactively trim models with n_ctx at or below this
+
 # ------------------------------------------------------------------
 # Globals
 # ------------------------------------------------------------------
@@ -707,10 +808,15 @@ class fetch:
         # Check error cache with lock protection
         async with _available_error_cache_lock:
             if endpoint in _available_error_cache:
-                if _is_fresh(_available_error_cache[endpoint], 300):
+                err_age = time.time() - _available_error_cache[endpoint]
-                    # Still within the short error TTL – pretend nothing is available
+                if err_age < 30:
+                    # Very fresh error (<30s) – endpoint likely still down, bail fast
                     return set()
-                # Error expired – remove it
+                elif err_age < 300:
+                    # Stale error (30-300s) – endpoint may have recovered, probe in background
+                    asyncio.create_task(fetch._refresh_available_models(endpoint, api_key))
+                    return set()
+                # Error expired (>300s) – remove and fall through to fresh fetch
                 del _available_error_cache[endpoint]
 
         # Request coalescing: check if another request is already fetching this endpoint
@@ -983,7 +1089,37 @@ async def _make_chat_request(model: str, messages: list, tools=None, stream: boo
     try:
         if use_openai:
             start_ts = time.perf_counter()
+            try:
                 response = await oclient.chat.completions.create(**params)
+            except Exception as e:
+                _e_str = str(e)
+                print(f"[_make_chat_request] caught {type(e).__name__}: {_e_str[:200]}")
+                if "exceed_context_size_error" in _e_str or "exceeds the available context size" in _e_str:
+                    err_body = getattr(e, "body", {}) or {}
+                    err_detail = err_body.get("error", {}) if isinstance(err_body, dict) else {}
+                    n_ctx_limit = err_detail.get("n_ctx", 0)
+                    actual_tokens = err_detail.get("n_prompt_tokens", 0)
+                    if not n_ctx_limit:
+                        raise
+                    msgs_to_trim = params.get("messages", [])
+                    cal_target = _calibrated_trim_target(msgs_to_trim, n_ctx_limit, actual_tokens)
+                    trimmed = _trim_messages_for_context(msgs_to_trim, n_ctx_limit, target_tokens=cal_target)
+                    print(f"[_make_chat_request] Context exceeded ({actual_tokens}/{n_ctx_limit} tokens, tiktoken_target={cal_target}), dropped {len(msgs_to_trim) - len(trimmed)} oldest message(s) and retrying")
+                    try:
+                        response = await oclient.chat.completions.create(**{**params, "messages": trimmed})
+                    except Exception as e2:
+                        if "exceed_context_size_error" in str(e2) or "exceeds the available context size" in str(e2):
+                            print(f"[_make_chat_request] Context still exceeded after trimming, also stripping tools")
+                            params_no_tools = {k: v for k, v in params.items() if k not in ("tools", "tool_choice")}
+                            response = await oclient.chat.completions.create(**{**params_no_tools, "messages": trimmed})
+                        else:
+                            raise
+                elif "image input is not supported" in _e_str:
+                    print(f"[_make_chat_request] Model {model} doesn't support images, retrying with text-only messages")
+                    params = {**params, "messages": _strip_images_from_messages(params.get("messages", []))}
+                    response = await oclient.chat.completions.create(**params)
+                else:
+                    raise
             if stream:
                 # For streaming, we need to collect all chunks
                 chunks = []
@@ -1212,6 +1348,22 @@ def transform_images_to_data_urls(message_list):
 
     return message_list
 
+def _strip_images_from_messages(messages: list) -> list:
+    """Remove image_url parts from message content, keeping only text."""
+    result = []
+    for msg in messages:
+        content = msg.get("content")
+        if isinstance(content, list):
+            text_only = [p for p in content if p.get("type") != "image_url"]
+            if len(text_only) == 1 and text_only[0].get("type") == "text":
+                content = text_only[0]["text"]
+            else:
+                content = text_only
+            result.append({**msg, "content": content})
+        else:
+            result.append(msg)
+    return result
+
 def _accumulate_openai_tc_delta(chunk, accumulator: dict) -> None:
     """Accumulate tool_call deltas from a single OpenAI streaming chunk.
 
@@ -1825,23 +1977,86 @@ async def chat_proxy(request: Request):
         oclient = openai.AsyncOpenAI(base_url=ep2base(endpoint), default_headers=default_headers, api_key=config.api_keys.get(endpoint, "no-key"))
     else:
         client = ollama.AsyncClient(host=endpoint)
+    # For OpenAI endpoints: make the API call in handler scope
+    # (try/except inside async generators is unreliable with Starlette's streaming)
+    start_ts = None
+    async_gen = None
+    if use_openai:
+        start_ts = time.perf_counter()
+        # Proactive trim: only for small-ctx models we've already seen run out of space
+        _lookup_model = _normalize_llama_model_name(model) if endpoint in config.llama_server_endpoints else model
+        _known_nctx = _endpoint_nctx.get((endpoint, _lookup_model))
+        if _known_nctx and _known_nctx <= _CTX_TRIM_SMALL_LIMIT:
+            _pre_target = int((_known_nctx - _known_nctx // 4) / 1.2)
+            _pre_est = _count_message_tokens(params.get("messages", []))
+            if _pre_est > _pre_target:
+                _pre_msgs = params.get("messages", [])
+                _pre_trimmed = _trim_messages_for_context(_pre_msgs, _known_nctx, target_tokens=_pre_target)
+                _dropped = len(_pre_msgs) - len(_pre_trimmed)
+                print(f"[ctx-pre] n_ctx={_known_nctx} est={_pre_est} target={_pre_target} dropped={_dropped}", flush=True)
+                params = {**params, "messages": _pre_trimmed}
+        try:
+            async_gen = await oclient.chat.completions.create(**params)
+        except Exception as e:
+            _e_str = str(e)
+            print(f"[chat_proxy] caught {type(e).__name__}: {_e_str[:200]}")
+            if "exceed_context_size_error" in _e_str or "exceeds the available context size" in _e_str:
+                err_body = getattr(e, "body", {}) or {}
+                err_detail = err_body.get("error", {}) if isinstance(err_body, dict) else {}
+                n_ctx_limit = err_detail.get("n_ctx", 0)
+                actual_tokens = err_detail.get("n_prompt_tokens", 0)
+                if not n_ctx_limit:
+                    await decrement_usage(endpoint, tracking_model)
+                    raise
+                if n_ctx_limit <= _CTX_TRIM_SMALL_LIMIT:
+                    _endpoint_nctx[(endpoint, model)] = n_ctx_limit
+                msgs_to_trim = params.get("messages", [])
+                cal_target = _calibrated_trim_target(msgs_to_trim, n_ctx_limit, actual_tokens)
+                trimmed = _trim_messages_for_context(msgs_to_trim, n_ctx_limit, target_tokens=cal_target)
+                print(f"[chat_proxy] Context exceeded ({actual_tokens}/{n_ctx_limit} tokens, tiktoken_target={cal_target}), dropped {len(msgs_to_trim) - len(trimmed)} oldest message(s) and retrying")
+                try:
+                    async_gen = await oclient.chat.completions.create(**{**params, "messages": trimmed})
+                except Exception as e2:
+                    _e2_str = str(e2)
+                    if "exceed_context_size_error" in _e2_str or "exceeds the available context size" in _e2_str:
+                        print(f"[chat_proxy] Context still exceeded after trimming messages, also stripping tools")
+                        params_no_tools = {k: v for k, v in params.items() if k not in ("tools", "tool_choice")}
+                        try:
+                            async_gen = await oclient.chat.completions.create(**{**params_no_tools, "messages": trimmed})
+                        except Exception:
+                            await decrement_usage(endpoint, tracking_model)
+                            raise
+                    else:
+                        await decrement_usage(endpoint, tracking_model)
+                        raise
+            elif "image input is not supported" in _e_str:
+                print(f"[chat_proxy] Model {model} doesn't support images, retrying with text-only messages")
+                try:
+                    params = {**params, "messages": _strip_images_from_messages(params.get("messages", []))}
+                    async_gen = await oclient.chat.completions.create(**params)
+                except Exception:
+                    await decrement_usage(endpoint, tracking_model)
+                    raise
+            else:
+                await decrement_usage(endpoint, tracking_model)
+                raise
+
     # 3. Async generator that streams chat data and decrements the counter
     async def stream_chat_response():
         try:
             # The chat method returns a generator of dicts (or GenerateResponse)
             if use_openai:
-                start_ts = time.perf_counter()
+                _async_gen = async_gen  # established in handler scope above
-                async_gen = await oclient.chat.completions.create(**params)
             else:
                 if opt == True:
                     # Use the dedicated MOE helper function
-                    async_gen = await _make_moe_requests(model, messages, tools, think, _format, options, keep_alive)
+                    _async_gen = await _make_moe_requests(model, messages, tools, think, _format, options, keep_alive)
                 else:
-                    async_gen = await client.chat(model=model, messages=messages, tools=tools, stream=stream, think=think, format=_format, options=options, keep_alive=keep_alive, logprobs=logprobs, top_logprobs=top_logprobs)
+                    _async_gen = await client.chat(model=model, messages=messages, tools=tools, stream=stream, think=think, format=_format, options=options, keep_alive=keep_alive, logprobs=logprobs, top_logprobs=top_logprobs)
             if stream == True:
                 tc_acc = {}  # accumulate OpenAI tool-call deltas across chunks
                 content_parts: list[str] = []
-                async for chunk in async_gen:
+                async for chunk in _async_gen:
                     if use_openai:
                         _accumulate_openai_tc_delta(chunk, tc_acc)
                         chunk = rechunk.openai_chat_completion2ollama(chunk, stream, start_ts)
@@ -1860,6 +2075,20 @@ async def chat_proxy(request: Request):
                     # Accumulate and store cache on done chunk — before yield so it always runs
                     # Works for both Ollama-native and OpenAI-compatible backends; chunks are
                     # already converted to Ollama format by rechunk before this point.
+                    if getattr(chunk, "done", False):
+                        # Detect context exhaustion mid-generation for small-ctx models
+                        _dr = getattr(chunk, "done_reason", None)
+                        # Only cache when no max_tokens limit was set — otherwise
+                        # finish_reason=length might just mean max_tokens was hit,
+                        # not that the context window was exhausted.
+                        _req_max_tok = params.get("max_tokens") or params.get("max_completion_tokens") or params.get("num_predict")
+                        if _dr == "length" and not _req_max_tok:
+                            _pt = getattr(chunk, "prompt_eval_count", 0) or 0
+                            _ct = getattr(chunk, "eval_count", 0) or 0
+                            _inferred_nctx = _pt + _ct
+                            if 0 < _inferred_nctx <= _CTX_TRIM_SMALL_LIMIT:
+                                _endpoint_nctx[(endpoint, model)] = _inferred_nctx
+                                print(f"[ctx-cache] done_reason=length → cached n_ctx={_inferred_nctx} for ({endpoint},{model})", flush=True)
                     if _cache is not None and not _is_moe and _cache_enabled:
                         if chunk.message and getattr(chunk.message, "content", None):
                             content_parts.append(chunk.message.content)
@@ -1884,18 +2113,18 @@ async def chat_proxy(request: Request):
                     yield json_line.encode("utf-8") + b"\n"
             else:
                 if use_openai:
-                    response = rechunk.openai_chat_completion2ollama(async_gen, stream, start_ts)
+                    response = rechunk.openai_chat_completion2ollama(_async_gen, stream, start_ts)
                     response = response.model_dump_json()
                 else:
-                    response = async_gen.model_dump_json()
+                    response = _async_gen.model_dump_json()
-                    prompt_tok = async_gen.prompt_eval_count or 0
+                    prompt_tok = _async_gen.prompt_eval_count or 0
-                    comp_tok   = async_gen.eval_count or 0
+                    comp_tok   = _async_gen.eval_count or 0
                     if prompt_tok != 0 or comp_tok != 0:
                         await token_queue.put((endpoint, tracking_model, prompt_tok, comp_tok))
                 json_line = (
                     response
-                    if hasattr(async_gen, "model_dump_json")
+                    if hasattr(_async_gen, "model_dump_json")
-                    else orjson.dumps(async_gen)
+                    else orjson.dumps(_async_gen)
                 )
                 cache_bytes = json_line.encode("utf-8") + b"\n"
                 yield cache_bytes
@@ -2604,9 +2833,13 @@ async def ps_details_proxy(request: Request):
             *[_fetch_llama_props(ep, mid) for ep, mid in props_requests]
         )
 
-        for model_dict, (n_ctx, is_sleeping) in zip(llama_models_pending, props_results):
+        for (ep, raw_id), model_dict, (n_ctx, is_sleeping) in zip(props_requests, llama_models_pending, props_results):
             if n_ctx is not None:
                 model_dict["context_length"] = n_ctx
+                if 0 < n_ctx <= _CTX_TRIM_SMALL_LIMIT:
+                    normalized = _normalize_llama_model_name(raw_id)
+                    _endpoint_nctx[(ep, normalized)] = n_ctx
+                    print(f"[ctx-cache/ps] cached n_ctx={n_ctx} for ({ep},{normalized})", flush=True)
             if not is_sleeping:
                 models.append(model_dict)
 
@@ -2686,6 +2919,21 @@ async def openai_embedding_proxy(request: Request):
         model = payload.get("model")
         doc = payload.get("input")
 
+        # Normalize multimodal input: extract only text parts for embedding models
+        if isinstance(doc, list):
+            normalized = []
+            for item in doc:
+                if isinstance(item, dict):
+                    # Multimodal content part - extract text only, skip images
+                    if item.get("type") == "text":
+                        normalized.append(item.get("text", ""))
+                    # Skip image_url and other non-text types
+                else:
+                    normalized.append(item)
+            doc = normalized if len(normalized) != 1 else normalized[0]
+        elif isinstance(doc, dict) and doc.get("type") == "text":
+            doc = doc.get("text", "")
+
         if not model:
             raise HTTPException(
                 status_code=400, detail="Missing required field 'model'"
@@ -2819,7 +3067,7 @@ async def openai_chat_completions_proxy(request: Request):
     endpoint, tracking_model = await choose_endpoint(model)
     base_url = ep2base(endpoint)
     oclient = openai.AsyncOpenAI(base_url=base_url, default_headers=default_headers, api_key=config.api_keys.get(endpoint, "no-key"))
-    # 3. Async generator that streams completions data and decrements the counter
+    # 3. Helpers and API call — done in handler scope so try/except works reliably
     async def _normalize_images_in_messages(msgs: list) -> list:
         """Fetch remote image URLs and convert them to base64 data URLs so
         Ollama/llama-server can handle them without making outbound HTTP requests."""
@@ -2854,25 +3102,95 @@ async def openai_chat_completions_proxy(request: Request):
             resolved.append({**msg, "content": new_content})
         return resolved
 
-    async def stream_ochat_response():
+    # Make the API call in handler scope — try/except inside async generators is unreliable
-        try:
+    # with Starlette's streaming machinery, so we resolve errors here before the generator starts.
-            # The chat method returns a generator of dicts (or GenerateResponse)
-            try:
-                # For non-external endpoints (Ollama, llama-server), resolve remote
-                # image URLs to base64 data URLs so the server can handle them locally.
     send_params = params
     if not is_ext_openai_endpoint(endpoint):
         resolved_msgs = await _normalize_images_in_messages(params.get("messages", []))
         send_params = {**params, "messages": resolved_msgs}
+    # Proactive trim: only for small-ctx models we've already seen run out of space
+    _lookup_model = _normalize_llama_model_name(model) if endpoint in config.llama_server_endpoints else model
+    _known_nctx = _endpoint_nctx.get((endpoint, _lookup_model))
+    if _known_nctx and _known_nctx <= _CTX_TRIM_SMALL_LIMIT:
+        _pre_target = int(((_known_nctx - _known_nctx // 4)) / 1.2)
+        _pre_est = _count_message_tokens(send_params.get("messages", []))
+        if _pre_est > _pre_target:
+            _pre_msgs = send_params.get("messages", [])
+            _pre_trimmed = _trim_messages_for_context(_pre_msgs, _known_nctx, target_tokens=_pre_target)
+            _dropped = len(_pre_msgs) - len(_pre_trimmed)
+            print(f"[ctx-pre] n_ctx={_known_nctx} est={_pre_est} target={_pre_target} dropped={_dropped}", flush=True)
+            send_params = {**send_params, "messages": _pre_trimmed}
+    try:
         async_gen = await oclient.chat.completions.create(**send_params)
-            except openai.BadRequestError as e:
+    except Exception as e:
-                # If tools are not supported by the model, retry without tools
+        _e_str = str(e)
-                if "does not support tools" in str(e):
+        _is_ctx_err = "exceed_context_size_error" in _e_str or "exceeds the available context size" in _e_str
-                    print(f"[openai_chat_completions_proxy] Model {model} doesn't support tools, retrying without tools")
+        print(f"[ochat] caught={type(e).__name__} ctx={_is_ctx_err} msg={_e_str[:120]}", flush=True)
+        if "does not support tools" in _e_str:
+            # Model doesn't support tools — retry without them
+            print(f"[ochat] retry: no tools", flush=True)
+            try:
                 params_without_tools = {k: v for k, v in send_params.items() if k != "tools"}
                 async_gen = await oclient.chat.completions.create(**params_without_tools)
-                else:
+            except Exception:
+                await decrement_usage(endpoint, tracking_model)
                 raise
+        elif _is_ctx_err:
+            # Backend context limit hit — apply sliding-window trim (context-shift at message level)
+            err_body = getattr(e, "body", {}) or {}
+            err_detail = err_body.get("error", {}) if isinstance(err_body, dict) else {}
+            n_ctx_limit = err_detail.get("n_ctx", 0)
+            actual_tokens = err_detail.get("n_prompt_tokens", 0)
+            print(f"[ctx-trim] n_ctx={n_ctx_limit} actual={actual_tokens}", flush=True)
+            if not n_ctx_limit:
+                await decrement_usage(endpoint, tracking_model)
+                raise
+            if n_ctx_limit <= _CTX_TRIM_SMALL_LIMIT:
+                _endpoint_nctx[(endpoint, model)] = n_ctx_limit
+
+            msgs_to_trim = send_params.get("messages", [])
+            try:
+                cal_target = _calibrated_trim_target(msgs_to_trim, n_ctx_limit, actual_tokens)
+                trimmed_messages = _trim_messages_for_context(msgs_to_trim, n_ctx_limit, target_tokens=cal_target)
+            except Exception as _helper_exc:
+                print(f"[ctx-trim] helper crash: {type(_helper_exc).__name__}: {str(_helper_exc)[:100]}", flush=True)
+                await decrement_usage(endpoint, tracking_model)
+                raise
+            dropped = len(msgs_to_trim) - len(trimmed_messages)
+            print(f"[ctx-trim] target={cal_target} dropped={dropped} remaining={len(trimmed_messages)} retrying-1", flush=True)
+            try:
+                async_gen = await oclient.chat.completions.create(**{**send_params, "messages": trimmed_messages})
+                print(f"[ctx-trim] retry-1 ok", flush=True)
+            except Exception as e2:
+                _e2_str = str(e2)
+                if "exceed_context_size_error" in _e2_str or "exceeds the available context size" in _e2_str:
+                    # Still too large — tool definitions likely consuming too many tokens, strip them too
+                    print(f"[ctx-trim] retry-1 still exceeded, stripping tools retrying-2", flush=True)
+                    params_no_tools = {k: v for k, v in send_params.items() if k not in ("tools", "tool_choice")}
+                    try:
+                        async_gen = await oclient.chat.completions.create(**{**params_no_tools, "messages": trimmed_messages})
+                        print(f"[ctx-trim] retry-2 ok", flush=True)
+                    except Exception:
+                        await decrement_usage(endpoint, tracking_model)
+                        raise
+                else:
+                    await decrement_usage(endpoint, tracking_model)
+                    raise
+        elif "image input is not supported" in _e_str:
+            # Model doesn't support images — strip and retry
+            print(f"[openai_chat_completions_proxy] Model {model} doesn't support images, retrying with text-only messages")
+            try:
+                async_gen = await oclient.chat.completions.create(**{**send_params, "messages": _strip_images_from_messages(send_params.get("messages", []))})
+            except Exception:
+                await decrement_usage(endpoint, tracking_model)
+                raise
+        else:
+            await decrement_usage(endpoint, tracking_model)
+            raise
+
+    # 4. Async generator — only streams the already-established async_gen
+    async def stream_ochat_response():
+        try:
             if stream == True:
                 content_parts: list[str] = []
                 usage_snapshot: dict = {}
@@ -2909,6 +3227,15 @@ async def openai_chat_completions_proxy(request: Request):
                             prompt_tok, comp_tok = llama_usage
                     if prompt_tok != 0 or comp_tok != 0:
                         await token_queue.put((endpoint, tracking_model, prompt_tok, comp_tok))
+                    # Detect context exhaustion mid-generation for small-ctx models.
+                    # Guard: skip if max_tokens was set in the request — finish_reason=length
+                    # could just mean the caller's token budget was exhausted, not the context window.
+                    _req_max_tok = send_params.get("max_tokens") or send_params.get("max_completion_tokens")
+                    if chunk.choices and chunk.choices[0].finish_reason == "length" and not _req_max_tok:
+                        _inferred_nctx = (prompt_tok + comp_tok) or 0
+                        if 0 < _inferred_nctx <= _CTX_TRIM_SMALL_LIMIT:
+                            _endpoint_nctx[(endpoint, model)] = _inferred_nctx
+                            print(f"[ctx-cache] finish_reason=length → cached n_ctx={_inferred_nctx} for ({endpoint},{model})", flush=True)
                 # Cache assembled streaming response — before [DONE] so it always runs
                 if _cache is not None and _cache_enabled and content_parts:
                     assembled = orjson.dumps({
@@ -3044,10 +3371,15 @@ async def openai_completions_proxy(request: Request):
     oclient = openai.AsyncOpenAI(base_url=base_url, default_headers=default_headers, api_key=config.api_keys.get(endpoint, "no-key"))
 
     # 3. Async generator that streams completions data and decrements the counter
+    # Make the API call in handler scope (try/except inside async generators is unreliable)
+    try:
+        async_gen = await oclient.completions.create(**params)
+    except Exception:
+        await decrement_usage(endpoint, tracking_model)
+        raise
+
     async def stream_ocompletions_response(model=model):
         try:
-            # The chat method returns a generator of dicts (or GenerateResponse)
-            async_gen = await oclient.completions.create(**params)
             if stream == True:
                 text_parts: list[str] = []
                 usage_snapshot: dict = {}