[ModelProvider] error. Reason: JSON parse error: [json.exception.parse_error.101] parse error at line 1, column 1: attempting to parse an empty input; check that your input string or stream contains the expected JSON

[zixiao-ye]

Version

Flock v0.5.0
DuckDB v1.4.2 (Andium) 68d7555f68

Behavior

When trying the getting started Ollama example with local model

CREATE SECRET (
    TYPE OLLAMA,
    API_URL '127.0.0.1:11434'
);

CREATE MODEL(
   'QuackingModel',
   'qwen3:4b',
   'ollama',
   {"tuple_format": "json", "batch_size": 32, "model_parameters": {"temperature": 0.7}}
);

SELECT llm_complete(
    {'model_name': 'QuackingModel'},
    {'prompt': 'Talk like a duck 🦆 and write a poem about a database 📚'}
);

I got the following error:

[ModelProvider] error. Reason: JSON parse error: [json.exception.parse_error.101] parse error at line 1, column 1: attempting to parse an empty input; check that your input string or stream contains the expected JSON

I have download the model and checked that Ollama is running locally on 127.0.0.1:11434.
Maybe do you know Why I got this error?

[anasdorbani]

Hi @zixiao-ye, could you try running the following query? It looks like the response format from Ollama might differ from what we expect, or it's not returning a valid response:

curl http://localhost:11434/api/generate \
  -H "Content-Type: application/json" \
  -d '{
        "model": "qwen3:4b",
        "prompt": "Talk like a duck 🦆 and write a poem about a database 📚",
        "stream": false,
        "options": {
          "temperature": 0.7
        }
      }'

Let us know what output you get so we can compare.

[zixiao-ye]

I got the following reply:

{"model":"qwen3:4b","created_at":"2025-11-21T15:17:16.770388Z","response":"*(Waddles in with a squishy hop, beak slightly open, quacking softly)*  \n\nQuack! Quack! My database’s deep,  \nWhere **sneaky data** sleeps!  \nRows like reeds, columns wide—  \nA **quack-quiet** tide inside.  \n\nI swim through **tables** (so clean!),  \nWith **queries** that glide between.  \nNo fish can hide—*all* are here,  \nIn **rows** and **columns**, **never** fear!  \n\n*Quack!* A **join**—a sudden splash!  \n*Quack!* A **where**—a gentle dash.  \n**Indexes**? Just **duck-tails** swish!  \n**Errors?** *Splish!* ...*no* mish!  \n\n**Data**? *Squish!* Like cold mud,  \n**Keys**? Like **wet** pebbles—*good* for mud!  \n**Update**? A **quack**-fast leap!  \n**Delete**? *Plop!* ...*just* a sleep.  \n\nSo if you **search** for **truth** in deep,  \n**Quack**—*this* is where I keep:  \n**Your** **database**... *squishy*, **real**,  \n**Duck**-**hearted**... **always** **whole**!  \n\n*(Waddles off, leaving a wet trail of \"SELECT * FROM JOY\")* 🦆💧📚","thinking":"Okay, the user wants me to talk like a duck and write a poem about a database. That's... oddly specific but super fun. \n\nFirst, I need to unpack the request. \"Talk like a duck\" means I should use duck sounds and movements - quacking, waddling, maybe even duck-related metaphors. And \"poem about a database\" is techy but poetic. User's probably tech-savvy but wants humor and creativity. Maybe they're a developer who's stressed about databases and wants a laugh? Or a teacher making tech fun for students? \n\nHmm... key points to hit:\n- Duck speech: \"Quack!\" \"Waddle!\" \"Squish!\" (gotta include those sounds)\n- Database features: tables, queries, data, rows/columns, connections\n- Must be playful but accurate enough for techies. No dumbing down - they'll spot if it's too childish.\n\n*Brainstorming lines...*\nStart with duck quacking about data... \"Quack! Quack! My database's deep\" - yeah, that sets the tone. Compare database to water? Ducks love water! \"Squishy rows\" for tables, \"query\" as duck swimming... \n\n*Potential pitfalls:*\nDon't make it too technical (no SQL syntax) but don't be vague either. User said \"poem\" so rhythm matters. Maybe AABB rhyme scheme? \n\n*User vibes check:* They used emojis (🦆📚) so they're playful. Probably wants to smile while reading. Might be testing if I can be creative with constraints. \n\n*Structure idea:*\n1. Opening: Duck quacking at database\n2. Database as water (duck habitat)\n3. Data as fish (metaphor)\n4. Query as duck swimming\n5. Close with duck's happy quack\n\n*Final check:* Should I add \"error messages\" as duck splashing? Yeah! \"Error? Splish! Splash!\" keeps it light. \n\n*Self-edit:* Avoid \"schema\" or \"indexes\" - too jargony. Keep it simple like \"sneaky data\" for hidden info. \n\n...Time to write. *Quacks mentally* Ready!\n","done":true,"done_reason":"stop","context":[151644,872,198,36210,1075,264,35985,11162,99,228,323,3270,264,32794,911,264,4625,11162,241,248,151645,198,151644,77091,198,151667,198,32313,11,279,1196,6801,752,311,3061,1075,264,35985,323,3270,264,32794,911,264,4625,13,2938,594,1112,73444,3151,714,2256,2464,13,4710,5338,11,358,1184,311,31065,279,1681,13,330,36210,1075,264,35985,1,3363,358,1265,990,35985,10362,323,19029,481,922,8985,11,289,718,2718,11,7196,1496,35985,13904,30668,1087,13,1597,330,5368,336,911,264,4625,1,374,13014,88,714,75434,13,2657,594,4658,13014,1331,402,13714,714,6801,27385,323,27597,13,10696,807,2299,264,15754,879,594,31547,911,31806,323,6801,264,12550,30,2521,264,11079,3259,13014,2464,369,4143,30,4710,80022,1112,1376,3501,311,4201,510,12,45770,8806,25,330,2183,473,8958,330,54,20137,8958,330,50,446,812,8958,320,70,22193,2924,1846,10362,340,12,9994,4419,25,12632,11,19556,11,821,11,6978,89274,82,11,13234,198,12,15465,387,56069,714,13382,3322,369,13014,550,13,2308,29255,287,1495,481,807,3278,7702,421,432,594,2238,93171,382,9,65077,26413,287,5128,1112,5618,3479,448,35985,922,8985,911,821,1112,330,2183,473,0,3406,473,0,3017,4625,594,5538,1,481,21639,11,429,7289,279,16232,13,23768,4625,311,3015,30,67782,2948,3015,0,330,50,446,812,88,6978,1,369,12632,11,330,1631,1,438,35985,23380,1112,4710,9,95611,80975,25,5618,8002,944,1281,432,2238,10916,320,2152,7870,19482,8,714,1513,944,387,39046,2987,13,2657,1053,330,5368,336,1,773,36290,12850,13,10696,88580,96934,12859,30,4710,9,1474,89849,1779,53386,2379,1483,99066,320,148735,144880,8,773,807,2299,56069,13,37154,6801,311,15289,1393,5290,13,33251,387,7497,421,358,646,387,11521,448,16982,13,4710,9,22952,4522,25,5618,16,13,40037,25,45770,922,8985,518,4625,198,17,13,9994,438,3015,320,72970,38546,340,18,13,2885,438,7640,320,4059,1342,269,340,19,13,11361,438,35985,23380,198,20,13,13032,448,35985,594,6247,922,473,271,9,19357,1779,53386,12260,358,912,330,841,6605,1,438,35985,12503,19049,30,21607,0,330,1454,30,51198,812,0,45442,8958,13598,432,3100,13,4710,9,12092,22142,53386,34006,330,17349,1,476,330,71227,1,481,2238,502,858,3549,13,13655,432,4285,1075,330,82,811,28100,821,1,369,8177,3546,13,4710,1112,1462,311,3270,13,353,2183,7821,33225,9,30982,4894,151668,271,6599,54,718,642,304,448,264,8167,812,88,7446,11,387,585,10078,1787,11,922,8985,56050,4806,18611,2183,473,0,3406,473,0,3017,4625,748,5538,11,2303,9064,3070,82,811,28100,821,334,71390,0,2303,9024,1075,312,6767,11,8147,6884,2293,2303,32,3070,446,473,12,43650,334,41938,4766,13,18611,40,16191,1526,3070,32110,334,320,704,4240,48643,2303,2354,3070,42835,334,429,85041,1948,13,2303,2753,7640,646,10265,2293,9,541,9,525,1588,11,2303,641,3070,1811,334,323,3070,16369,97219,3070,36493,334,8679,0,18611,9,2183,473,68222,362,3070,5987,334,2293,64,10968,34632,0,2303,9,2183,473,68222,362,3070,2870,334,2293,64,21700,23911,13,2303,334,62229,334,30,4599,3070,72970,2385,6209,334,2021,812,0,2303,334,13877,30,334,353,94159,812,68222,2503,9,2152,9,63396,0,18611,334,1043,334,30,353,50,446,812,68222,8909,9255,26183,11,2303,334,8850,334,30,8909,3070,86,295,334,1051,6066,642,2293,9,18536,9,369,26183,0,2303,334,4289,334,30,362,3070,446,473,334,12,9349,31471,0,2303,334,6435,334,30,353,47,22288,68222,2503,9,4250,9,264,6084,13,18611,4416,421,498,3070,1836,334,369,3070,58577,334,304,5538,11,2303,334,2183,473,334,2293,9,574,9,374,1380,358,2506,25,2303,334,7771,334,3070,12216,334,13,496,353,97407,812,88,12314,3070,7951,97219,2303,334,35,1942,334,12,334,17735,291,334,13,496,3070,32122,334,3070,66633,334,0,18611,6599,54,718,642,1007,11,9380,264,14401,8849,315,330,4858,353,4295,10247,56,899,9,11162,99,228,145825,144880],"total_duration":13168020208,"load_duration":1033643833,"prompt_eval_count":27,"prompt_eval_duration":153932083,"eval_count":776,"eval_duration":11848395507}%

Is this due to the fact that Flock parses replies differently from Ollama's current reply format? Is it done by the model_manager?