Spark-X2.6-4B's CRAZY Thinking Dial: 180 Tests on an 8GB GPU

Spark-X2.6-4B thinking modes tested: none vs low vs high, 180 graded runs on an 8 GB RTX 4060 laptop GPU. Does turning the dial up earn its tokens? In my tests: default to low. Every number and every failure is in the video. What I found (my tests, Q4_K_M, temperature 1.0, 20 challenges x 3 modes x 3 passes): - Correct out of 60: none 48, low 54, high 53 - Tokens generated: none 12,179, low 47,789, high 69,338 - Thinking fixed the logic puzzles (8 to 15 of 15) but did not help debugging, and high used about 45% more tokens than low for no better score - The default Ollama tag (8.2 GB) ran 49% on the CPU at about 8 tokens/s on my 8 GB card; the official Q4_K_M quant (3.5 GB) ran at about 60 tokens/s Chapters: 0:00 50 or 5? The bat-and-ball shock 0:54 What Spark-X2.6-4B is 1:46 Will it fit on an 8 GB GPU? 2:54 How the none / low / high dial works 3:30 The test: 20 challenges, 180 runs 4:17 Results: score, tokens and time 5:56 Results by category 7:15 The failures, word for word 8:31 What thinking costs 8:48 Using the dial in Ollama 9:29 Verdict and limits Links from the video: Model card: https://huggingface.co/XHToken/Spark-X2.6-4B Official GGUF files: https://huggingface.co/XHToken/Spark-X2.6-4B-GGUF Ollama page: https://ollama.com/SparkLLM/Spark-X2.6-4B Caveats: one 4-bit quantization, 20 tests I wrote myself, 3 runs each, one strict grader. The vendor benchmark charts are the developer's claims and I did not reproduce them. The repository date is repository history, not an official release date. If you are looking for cheap and reliable GPUs, I use RunPod: https://get.runpod.io/pe48 Disclosure: the narration is a clone of my own voice, and the thumbnail portrait is AI-generated from my photo. All numbers on screen come from the logged runs. Get GPUs Runpod: https://get.runpod.io/pe48 Get Hostinger: https://www.hostg.xyz/SHJiu Coupon Code "PROMPT" 10% discount 🔗 Connect with me: Email → [email protected] Ko-fi → https://ko-fi.com/promptengineer Patreon → https://www.patreon.com/PromptEngineer975 Book a Call → https://calendly.com/prompt-engineer48/call GitHub → https://github.com/PromptEngineer48 Twitter/X → https://twitter.com/prompt48 Website → https://promptengineer48.com/