Self-Training Agents: Hermes Agent, HF Traces, Skills, MCP & Finetuning — Merve Noyan, Hugging Face

What was stored

cues

201

whisperx 201

chunks

33

from 201 cues

keyframes

64

kept of 71 captured

frames with text

64

2,021 lines read

chapters

15

from the source metadata

keyframe bytes

8.2 MB

word timings on 201 cues

Provenance

Frames, and what the machine read

• 

  0:03 #0 done2 line(s)

  shot 0·sharpness 658.4

  1. AlEngineer0.98
  2. EUROPE1.00
• 

  0:08 #1 done2 line(s)

  shot 1·sharpness 829.1

  1. PRESENTINGSPONSOR1.00
  2. Google DeepMind1.00
• 

  0:13 #2 done3 line(s)

  shot 2·sharpness 907.7

  1. PLATINUM SPONSORS0.98
  2. # Braintrust0.96
  3. WorkOS OpenAI0.95
• 

  0:21 #3 done15 line(s)

  shot 3·sharpness 1494.4

  1. eer0.96
  2. raintrus1.00
  3. ze1.00
  4. jineer0.97
  5. Open Agent1.00
  6. OPE0.98
  7. eer1.00
  8. gger.de1.00
  9. Ecosystem1.00
  10. odal0.99
  11. Engineer1.00
  12. (and having an Intern Al Engineer at your fingertips)0.99
  13. EUROPE1.00
  14. AlEngineer0.98
  15. EUROPE1.00
• 

  0:24 #4 done12 line(s)

  shot 4·sharpness 771.5

  1. eer0.97
  2. aintrus1.00
  3. ze1.00
  4. ineer1.00
  5. Open/source1.00
  6. eer1.00
  7. ger.de1.00
  8. odal0.97
  9. ngineer1.00
  10. EUROPE1.00
  11. AlEngineer0.98
  12. EUROPE1.00
• 

  0:36 #5 done13 line(s)

  shot 5·sharpness 770.0

  1. er1.00
  2. raintrus1.00
  3. ze1.00
  4. jineer0.92
  5. PE0.99
  6. Open/source1.00
  7. eer0.96
  8. 1ger.de0.99
  9. dal1.00
  10. jineer1.00
  11. OPE1.00
  12. AlEngineer0.99
  13. EUROPE1.00
• 

  0:54 #6 done36 line(s)

  shot 6·sharpness 1277.7

  1. B1.00
  2. trust1.00
  3. Open-source1.00
  4. deepseek-ai/DeepSeek-V3.1-Terminuslike291Follow DeepSeek97.8k0.98
  5. Text GenerationTransformersSafetensors deepseek_v30.99
  6. conversational0.98
  7. custom_code1.00
  8. text-generation-inference1.00
  9. License: mit0.98
  10. Model card1.00
  11. Files and versions o xet0.91
  12. Community1.00
  13. ¿ Edit model card0.97
  14. dev1.00
  15. DeepSeek-V3.1-Terminus1.00
  16. Downloads last mont0.99
  17. 11,0701.00
  18. deepseek1.00
  19. Safetensors0.97
  20. Model size 685B params Tensortype BF16·F8_E4M3 · F320.94
  21. 0 Chat template Files info0.93
  22. er1.00
  23. Inference Providers1.00
  24. Novita1.00
  25. Hugging Fac0.98
  26. X Twitter0.93
  27. deepseek al0.97
  28. Text Generation1.00
  29. Examples1.00
  30. Input a message to start chatting with deepseek-ai/DeepSeek-V3.1-Terminus.1.00
  31. Introduction1.00
  32. This update maintains the model's original capabilities while addressing issues1.00
  33. reported by users, including:1.00
  34. I anmuaae rnncictanru- Darlurina inetanree nf mivad Chinaca.Fnalich tovt and0.89
  35. AlEngineer0.97
  36. EUROPE1.00
• 

  1:10 #7 done43 line(s)

  shot 7·sharpness 1485.0

  1. Open-source1.00
  2. deepseek-ai/DeepSeek-V3.1-Terminuslike2911.00
  3. Follow DeepSeek 97.8k0.98
  4. Text Generation1.00
  5. Transformers1.00
  6. Safetensors1.00
  7. deepseek_v31.00
  8. conversational1.00
  9. custom_code1.00
  10. text-generation-inference1.00
  11. License: mit1.00
  12. Model card1.00
  13. Files and versionsxet1.00
  14. Community 100.99
  15. Edit model card0.98
  16. 三0.69
  17. DeepSeek-V3.1-Terminus1.00
  18. Downloads last month1.00
  19. 11,0701.00
  20. deepseek1.00
  21. Safetensors①0.96
  22. Model size 6858 params0.98
  23. Tensor type BF16 · F8_E4M3 · F320.96
  24. φ Chat template0.95
  25. Files info0.99
  26. Inference ProvidersNEw0.98
  27. Novita1.00
  28. DeepSeek Homepage1.00
  29. Chat DeepSeek V31.00
  30. Hugging Face1.00
  31. DeepSeek AI0.97
  32. Text Generation0.98
  33. Examples1.00
  34. Discord DeepSeek AI1.00
  35. WeChat DeepSeek AI0.96
  36. X Twitter0.96
  37. deepseek ai0.96
  38. License MIT0.99
  39. Input a message to start chatting with deepseek-ai/DeepSeek-V3.1-Terminus.1.00
  40. Introduction1.00
  41. This update maintains the model's original capabilities while addressing issues1.00
  42. reported by users, including:1.00
  43. I anauaae roncictancu Deducina inctancec nf mived Chinece-Fnalich tovt and0.89
• 

  1:38 #8 done11 line(s)

  shot 8·sharpness 2987.7

  1. onAI0.83
  2. Why does open-source matter?1.00
  3. • Absolute control over models1.00
  4. • Cost reduction in multipliers1.00
  5. • Further customize, shrink models depending on your needs0.98
  6. • Guaranteed privacy for the end-user, enable on-0.99
  7. device/in-browser uses, data doesn't go to another server0.99
  8. NTRY1.00
  9. lind0.83
  10. AlEngineer0.98
  11. EUROPE1.00
• 

  1:59 #9 done8 line(s)

  shot 9·sharpness 2896.1

  1. Why does open-source matter?1.00
  2. • Absolute control over models1.00
  3. • Cost reduction in multipliers1.00
  4. • Further customize, shrink models depending on your needs0.99
  5. • Guaranteed privacy for the end-user, enable on-0.99
  6. device/in-browser uses, data doesn't go to another server1.00
  7. AlEngineer0.98
  8. EUROPE1.00
• 

  2:23 #10 done94 line(s)

  shot 10·sharpness 2730.2

  1. Artificial Analysis Index0.98
  2. Artificial Analysis Intelligence Index by Open Weights / Proprietary1.00
  3. 28 of 472 models1.00
  4. 00.67
  5. Artificial Analysis Intelligence Index v4.0 incorporates 10 evaluations: GDPval-AA, τ2-Bench Telecom, Terminal-0.99
  6. + Add model from specific provider0.97
  7. Bench Hard, SciCode, AA-LCR, AA-Omniscience, IFBench, Humanity's Last Exam, GPQA Diamond, CritPt1.00
  8. Proprietary1.00
  9. Open Weights0.96
  10. A Artificial Analysis0.99
  11. 571.00
  12. 571.00
  13. 531.00
  14. 521.00
  15. 521.00
  16. 511.00
  17. 501.00
  18. 501.00
  19. 491.00
  20. 491.00
  21. 481.00
  22. 471.00
  23. 461.00
  24. 451.00
  25. 421.00
  26. 391.00
  27. 371.00
  28. 361.00
  29. 361.00
  30. 341.00
  31. 331.00
  32. 321.00
  33. 271.00
  34. 261.00
  35. 241.00
  36. 241.00
  37. 181.00
  38. G1.00
  39. AI0.91
  40. 80.64
  41. AI0.93
  42. Z0.87
  43. a0.51
  44. K0.93
  45. G1.00
  46. a0.67
  47. G1.00
  48. AI0.93
  49. G1.00
  50. M0.79
  51. 80.83
  52. C0.93
  53. C0.94
  54. C0.95
  55. C0.71
  56. Ca0.59
  57. C0.94
  58. C0.86
  59. C0.93
  60. C0.97
  61. C.0.61
  62. C0.95
  63. C0.96
  64. C0.97
  65. C0.88
  66. Cn0.58
  67. C0.95
  68. (max)0.84
  69. Gemini 3.11.00
  70. Pro Preview0.99
  71. GPT-5.40.99
  72. Claude Opus0.96
  73. 4.6 (max)0.98
  74. Claude1.00
  75. GLM-5.190.99
  76. MiniMax-M2.7 ?0.93
  77. Grok 4.200.97
  78. 0309 v20.84
  79. GPT-5.4 mini0.96
  80. (xhigh)0.93
  81. Kimi K2.5 90.92
  82. Gemini 30.98
  83. Flash0.99
  84. Qwen3.5 397B0.98
  85. Haiku1.00
  86. Nova 2.0 Pro1.00
  87. Preview1.00
  88. (medium)0.97
  89. Preview1.00
  90. gpt-oss-120B1.00
  91. K-EXAONE ?0.91
  92. gpt-oss-20B1.00
  93. K2 Think V2 90.89
  94. Llama 4 Maverick1.00
• 

  2:35 #11 done57 line(s)

  shot 11·sharpness 2802.0

  1. Hugging Face Hub1.00
  2. Home for the open-source machine learning community: share & discover models,0.99
  3. datasets, apps, connect with the community and more!1.00
  4. Hugging Face1.00
  5. O Search models, datasets, users...0.96
  6. Models1.00
  7. Datasets1.00
  8. Spaces1.00
  9. Community0.97
  10. Docs Pricing0.99
  11. i0.56
  12. + New0.98
  13. Following3551.00
  14. New Post0.99
  15. Trending1.00
  16. last 7 days0.99
  17. Models Datasets0.98
  18. Spaces1.00
  19. Papers Collections Community1.00
  20. Posts1.00
  21. Models1.00
  22. Datasets1.00
  23. Spaces1.00
  24. merve1.00
  25. Upvotes Likes Articles0.99
  26. Profile1.00
  27. moonshotai/Kimi-K2-Instruct1.00
  28. Inbox (7113)1.00
  29. hysts updated a Space 2 minutes ago0.98
  30. Text GenerationUpdated...25.1k1.23k0.99
  31. Settings1.00
  32. $ Billing0.99
  33. SeqTex0.94
  34. DeepSite v20.99
  35. 10.2k1.00
  36. SeqTex generates texture based on textual conditions1.00
  37. Generate any application with DeepSeek0.99
  38. Organizations1.00
  39. Hugging Face0.99
  40. Wauplin updated a Space5 minutes ago1.00
  41. HuggingFaceTB/SmolLM3-3B1.00
  42. Text Generation3BUp...59.5k4850.97
  43. G Google0.98
  44. Responses.js1.00
  45. 090.83
  46. Deprem Yapay Zeka1.00
  47. Check out https://github.com/huggingface/responses.js0.98
  48. mistralai/Devstral-Small-25071.00
  49. Notebooks-explorers1.00
  50. Text Generation 24BU...9.58k2330.95
  51. - SODA0.89
  52. — PyTorch Image Models0.95
  53. Weyaxi updated a dataset 5 minutes ago0.98
  54. black-forest-labs/FLUX.1-Kontext-dev0.99
  55. Image-to-Image·Updated ...274k1.68k0.95
  56. Templates1.00
  57. Weyaxi/huggingface-leaderboard1.00
• 

  2:50 #12 done48 line(s)

  shot 12·sharpness 1283.1

  1. Hugging Face Hub1.00
  2. Home for the machine learning community: share & discover1.00
  3. mo1.00
  4. 2.7M models of all libraries and tasks1.00
  5. more!1.00
  6. 925k+ datasets1.00
  7. + Ne0.94
  8. Spaces (apps)1.00
  9. New Post1.00
  10. Trending1.00
  11. last 7 days1.00
  12. Collections1.00
  13. Community1.00
  14. Posts1.00
  15. AIl0.53
  16. Models1.00
  17. Datasets1.00
  18. Spaces1.00
  19. merve0.98
  20. Prof1.00
  21. ecosystem: open-source libraries1.00
  22. cruct0.98
  23. Inbo0.98
  24. 25.1k1.00
  25. Setti0.99
  26. $ Billir0.92
  27. (transformers, diffusers & more!)1.00
  28. 10.2k1.00
  29. ith DeepSeek1.00
  30. Organiza1.00
  31. HuggingFaceTB/SmolLM3-3B1.00
  32. Hugging Face1.00
  33. Wauplin updated a Space5 minutes ago0.97
  34. Text Generation 3B·Up...59.5k0.96
  35. 4851.00
  36. G Google1.00
  37. Responses.js1.00
  38. Deprem Yapay Zeka0.97
  39. Check out https://github.com/huggingface/responses.js0.99
  40. mistralai/Devstral-Small-25071.00
  41. Notebooks-explorers1.00
  42. - SODA0.88
  43. —PyTorch Image Models0.97
  44. Weyaxi updated a dataset 5 minutes ago0.98
  45. black-forest-labs/FLUX.1-Kontext-dev0.99
  46. Templates1.00
  47. Weyaxi/huggingface-leaderboard1.00
  48. Vorse0.70
• 

  3:07 #13 done53 line(s)

  shot 13·sharpness 1126.2

  1. /models1.00
  2. Main1.00
  3. Tasks1.00
  4. Libraries1.00
  5. Languages Licenses0.97
  6. Models2,665,3241.00
  7. Filter by name0.98
  8. Full-text search0.99
  9. Inference Available0.99
  10. ↑↓ Sort: Trending1.00
  11. Other1.00
  12. Tasks1.00
  13. Image-Text-to-Text·: 36B·Updated about 3 hours ago·259k·6040.94
  14. Qwen/Qwen3.5-35B-A3B1.00
  15. Text Generation0.99
  16. Any-to-Any0.96
  17. P0.61
  18. Image-Text-to-Text1.00
  19. Image-to-Text1.00
  20. Qwen/Qwen3.5-27B1.00
  21. Image-Text-to-Text28B·Updated 2 days ago108k3960.98
  22. 图0.72
  23. Image-to-Image1.00
  24. Text-to-Image0.98
  25. Text-to-Video1.00
  26. Text-to-Speech1.00
  27. +441.00
  28. Qwen/Qwen3.5-397B-A17B1.00
  29. Image-Text-to-Text 403B·Updated 4 days ago·726k1.11k0.96
  30. Parameters1.00
  31. <1B0.99
  32. 6B0.96
  33. 12B0.99
  34. 32B1.00
  35. 128B0.90
  36. >500B0.99
  37. Qwen/Qwen3.5-122B-A10B1.00
  38. Image-Text-to-Text · 125B · Updated 3 days ago·108k· 3240.91
  39. Libraries1.00
  40. unsloth/Qwen3.5-35B-A3B-GGUF1.00
  41. PyTorch0.97
  42. TensorFlow1.00
  43. X JAX0.82
  44. Image-Text-to-Text 35B·Updated 3 days ago·265k·2750.94
  45. Transformers1.00
  46. Diffusers1.00
  47. zai-org/GLM-51.00
  48. sentence-transformers1.00
  49. Safetensors1.00
  50. Text Generation754B·Updated 14 days ago189k1.63k0.96
  51. ONNX1.00
  52. GGUF1.00
  53. Transformers.js0.98
• 

  3:25 #14 done8 line(s)

  shot 14·sharpness 3036.3

  1. models → agents, serve locally1.00
  2. Agentic LLMs (thinking + tool calling): gpt-oss, Gemma-4,0.99
  3. Minimax M2.7, GLM-5, Nemotron3-Super0.99
  4. Agentic vision models (thinking + CUA): Qwen3.50.99
  5. (Alibaba), Kimi-K2.50.98
  6. mlx_lm.generate --prompt "How tall is Mt Everest?"0.99
  7. vllm serve Qwen/Qwen3-8B # then query with OpenAI Completion1.00
  8. llama-server -m model.gguf --port 80801.00
• 

  3:53 #15 skipped

  shot 15·duplicate of #14

• 

  4:34 #16 done67 line(s)

  shot 16·sharpness 1361.1

  1. compare open models1.00
  2. Main1.00
  3. Tasks1.00
  4. Libraries1.00
  5. Languages Licenses Other1.00
  6. Datasets 171.00
  7. Filter by name1.00
  8. Full-text search1.00
  9. ↑↓ Sort: Trending0.98
  10. Modalities0.99
  11. openai/gsm8k1.00
  12. allenai/olmOCR-bench1.00
  13. 3D1.00
  14. Audio1.00
  15. Document1.00
  16. Geospatial1.00
  17. Benchmark·Updated 17 days ago·17.6k·±765k1.24k0.96
  18. Benchmark · Updated Feb 19·± 4.14k · 1760.90
  19. Image1.00
  20. Tabular1.00
  21. Text1.00
  22. Time-series1.00
  23. ScaleAI/SWE-bench_Pro1.00
  24. cais/hle1.00
  25. Video1.00
  26. Benchmark·Updated Feb 23·731·±692k·780.91
  27. Benchmark · Updated Jan 20· 2.5k·±46.1k ·7610.91
  28. Size (rows)1.00
  29. SWE-bench/SWE-bench_Verified1.00
  30. collinear-ai/yc-bench1.00
  31. <1K0.93
  32. >1T0.95
  33. Benchmark·Updated Feb 27· 500·126k·280.92
  34. Benchmark - Updated 17 days ago · ± 104 · 150.91
  35. TIGER-Lab/MMLU-Pro1.00
  36. MathArena/aime_20261.00
  37. Format0.97
  38. Benchmark · Updated 29 days ago·12.1k ·± 113k·4640.92
  39. Benchmark · Updated Feb 16 · 30± 12.7k · 280.90
  40. 4 json0.87
  41. EE CSV0.78
  42. parquet0.94
  43. optimized-parquet1.00
  44. imagefolder1.00
  45. soundfolder1.00
  46. webdataset1.00
  47. harborframework/terminal-bench-2.01.00
  48. Benchmark · Updated Feb 17 · ± 2.98k - 190.89
  49. Idavidrein/gpqa1.00
  50. Benchmark · Updated Mar 5· 1.25k ·± 105k · 4080.87
  51. text0.94
  52. arrow0.98
  53. mteb/arguana1.00
  54. FutureMa/EvasionBench1.00
  55. Type1.00
  56. Benchmark·Updated Feb 22·11.5k-± 12.8k·50.91
  57. Benchmark · Updated Feb 19·16.7k·± 177·850.92
  58. Benchmark×1.00
  59. è Traces0.94
  60. mteb/BRIGHT1.00
  61. hf-audio/open-asr-leaderboard1.00
  62. Benchmark · Updated 7 days ago·1.35M - ±662-20.90
  63. Benchmark - Updated 6 days ago · 99.5k · ± 20.1k - 70.90
  64. likaixin/ScreenSpot-Pro1.00
  65. nvidia/compute-eval1.00
  66. Benchmark - Updated 22 days ago ·± 7.73k - 600.91
  67. Benchmark - Updated 20 days ago · 2.46k - ± 5.63k · 220.89
• 

  4:53 #17 done45 line(s)

  shot 17·sharpness 1624.0

  1. compare open models1.00
  2. Datasets:ScaleAl/SWE-bench_Pro1.00
  3. like1.00
  4. 781.00
  5. FollowScale Al0.98
  6. 2841.00
  7. Benchmark1.00
  8. Modalities:1.00
  9. Text1.00
  10. Formats:1.00
  11. parquet0.95
  12. Size:1.00
  13. <1K1.00
  14. Libraries:1.00
  15. Datasets1.00
  16. pandas1.00
  17. Polars1.00
  18. +11.00
  19. Dataset card1.00
  20. 田Data Studio1.00
  21. Files and versions xet0.96
  22. Community1.00
  23. Leaderboard Official Benchmark0.97
  24. ① Learn more0.98
  25. Experimental1.00
  26. Task: SWE Bench Pro0.97
  27. #1.00
  28. MODEL1.00
  29. SCORE1.00
  30. zai-org/GLM-5.10.97
  31. 58.4*1.00
  32. 21.00
  33. MiniMaxAI/MiniMax-M2.51.00
  34. 55.41.00
  35. 31.00
  36. moonshotai/Kimi-K2.51.00
  37. 50.71.00
  38. Qwen/Qwen3-Coder-Next1.00
  39. source1.00
  40. 44.3*1.00
  41. 51.00
  42. Qwen/Qwen3-Coder-480B-A35B-Instruct0.99
  43. source1.00
  44. 38.71.00
  45. Show all 14 models0.99
• 

  5:22 #18 done35 line(s)

  shot 18·sharpness 1258.2

  1. Inference Providers for Qwen/Qwen3-VL-235B-A22B-Thinking0.99
  2. ×0.80
  3. Novita1.00
  4. ★ Auto0.89
  5. mervenoyan1.00
  6. t routing0.98
  7. Python1.00
  8. JavaScript1.00
  9. CcURL0.95
  10. huggingface_hub0.98
  11. requests openai1.00
  12. Stream1.00
  13. import os0.98
  14. Copy1.00
  15. from huggingface_hub import InferenceClient1.00
  16. client = InferenceClient(0.99
  17. vibe-check1.00
  18. provider="novita",1.00
  19. api_key=os.environ["HF_TOKEN"],0.99
  20. completion = client.chat.completions.create(1.00
  21. model="Qwen/Qwen3-VL-235B-A22B-Thinking",1.00
  22. go serverless1.00
  23. messages=[1.00
  24. {0.85
  25. "role": "user",0.99
  26. "content": [0.96
  27. {0.95
  28. "type": "text",0.98
  29. "text": "Describe this image in one sentence."1.00
  30. 3,0.64
  31. {0.92
  32. "type": "image_url",0.99
  33. "image_url": {0.99
  34. "url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Islan1.00
  35. print(completion.choices[0].message)1.00
• 

  5:35 #19 done48 line(s)

  shot 19·sharpness 2413.3

  1. compare providers across models0.99
  2. Inference Providers ·Metrics for top trending models0.98
  3. Browse all0.96
  4. Filter by model or provider...1.00
  5. Model0.98
  6. Provider1.00
  7. Input $/1M0.98
  8. Output $/1M0.99
  9. Context1.00
  10. Latency(s)1.00
  11. Throughput(t/s)0.99
  12. Ggoogle/gemma-4-31B-it0.99
  13. novita1.00
  14. $0.141.00
  15. $0.401.00
  16. 262,1441.00
  17. 1.301.00
  18. 301.00
  19. G google/gemma-4-26B-A4B-it0.98
  20. novita1.00
  21. $0.131.00
  22. $0.401.00
  23. 262,1441.00
  24. 2.961.00
  25. 131.00
  26. Qwen/Qwen3.5-9B0.97
  27. 370.87
  28. together1.00
  29. $0.101.00
  30. $0.151.00
  31. 262,1441.00
  32. 0.791.00
  33. 621.00
  34. zai-org/GLM-51.00
  35. 70.73
  36. novita cheapest0.97
  37. $1.001.00
  38. $3.201.00
  39. 202,8001.00
  40. 2.411.00
  41. 340.97
  42. zai-org/GLM-51.00
  43. 70.68
  44. together1.00
  45. $1.001.00
  46. $3.201.00
  47. 202,7521.00
  48. 0.611.00
• 

  6:14 #20 done

  shot 20·sharpness 2969.2

• 

  6:21 #21 skipped

  shot 21·duplicate of #20

• 

  6:55 #22 done

  shot 22·sharpness 948.8

• 

  7:20 #23 done

  shot 23·sharpness 2543.9

The page's on-screen-text budget of 600 lines is spent, so the last cards in this grid list fewer lines than they hold. Narrow the page with ?frames= to read them.

Transcript

201 cues· 2,780 words· 14,580 chars

Chapters

1. 0:00 Introduction to Open Agent Ecosystem
2. 0:39 Importance of Open Source in Machine Learning
3. 2:36 Hugging Face Hub overview
4. 3:06 Agentic models and Vision-LMs
5. 4:24 Benchmark datasets and model filtering
6. 5:16 Inference providers and model routing
7. 6:50 Local coding agents and tools
8. 7:46 Hermes agents for memory management
9. 9:20 Traces repository for agent sessions
10. 10:22 Tips for finding and serving local models
11. 12:07 Supercharging agents with Hugging Face skills
12. 13:41 Live demonstration of agent-driven fine-tuning
13. 14:41 Training vision models (object detection/segmentation)
14. 15:00 Using Model Context Protocol (MCP) for agents
15. 16:30 Case study: OCR processing for AI papers