LOADING DATASHEET
GROUNDED IN REAL HUMAN CONVERSATIONSYOUR BENCH
Gsm8k
★ NEW◆ OPEN
by Predibase
OPEN WEIGHTSLOCALSMALLEUROPE-MADE
RANKED #220 OF 267 ASSISTANTS · OVERALL #350 OF 6,544 · VIBE SCORE 5.3 · 37 VOICES
gsm8k on Hugging Face (text generation). 2,000 downloads. Open weights for local or hosted use.
HAPPY5.0
37 mentions
SAD-
0 mentions
VIBE TRAJECTORY
26 weeks · 46 voices
TOO QUIET
Not enough weekly voices to call a trend yet.
HOW TO USE IT
Open weights. You can use a hosted service, or download it and run it yourself, free.
RUN IT ON YOUR OWN MACHINE
Open weights: download Gsm8k once and it's yours. No subscription, no rate limits, works offline.
7B base parameters (Mistral-7B-v0.1) with a LoRA adapter, a Q4 quantized export is about 4.5 GB
- BARE MINIMUM
NVIDIA GeForce RTX 3060 12GB
12 GB VRAM
~$260 USEDQ4_K_M merged GGUF or 4-bit PEFT LoRA, ~45 tok/s, 4k context
- SWEET SPOT
Apple MacBook Air M3 16GB
16 GB UNIFIED RAM
~$850 REFURBISHEDQ8 / FP16 merged GGUF or full 8k context LoRA runtime, ~30 tok/s, 8k context
- FULL POWER
NVIDIA GeForce RTX 4070 Ti Super 16GB
16 GB VRAM
~$750 NEWFP16 unquantized base plus LoRA via vLLM or Hugging Face PEFT, ~75 tok/s, 8k context
LEARN HOW TO SET IT UP
WILL MY BOX RUN GSM8K?
The three cards above are researched picks. Search the machine you actually have — we only say yes if it should reply at a usable speed, not just load the weights and crawl.
HARDWARE PICKS & PRICES ARE RESEARCHED FROM THE LIVE WEB AND REFRESHED AUTOMATICALLY. TREAT THEM AS BALLPARK, NOT GOSPEL.
WHAT CAN YOU ACTUALLY DO WITH IT?
Everyday jobs on each of those rigs: how long Gsm8k takes, and how good the crowd says it is at that kind of work.
| THE JOB | BARE MINIMUMNVIDIA GeForce RTX 3060 12GB | SWEET SPOTApple MacBook Air M3 16GB | FULL POWERNVIDIA GeForce RTX 4070 Ti Super 16GB |
|---|---|---|---|
| Write an emaila paragraph or two, drafted from a one-line brief | ~11 S★★★★★ | ~17 S★★★★★ | ~7 S★★★★★ |
| Summarize a documenta long report boiled down to the points that matter | ~33 S★★★★★ | ~50 S★★★★★ | ~20 S★★★★★ |
| Build a websitea small landing page, markup and styles together | ~3 MIN★★★★★ | ~4.4 MIN★★★★★ | ~1.8 MIN★★★★★ |
| Build a backendan API with routes, storage and tests | ~9.3 MIN★★★★★ | ~14 MIN★★★★★ | ~5.6 MIN★★★★★ |
| Build a gamea playable browser game in one file | ~5.6 MIN★★★★★ | ~8.3 MIN★★★★★ | ~3.3 MIN★★★★★ |
STARS ARE THE COMMUNITY SCORE FOR THAT KIND OF WORK, DOCKED FOR HOW SQUASHED THE WEIGHTS GET AT EACH TIER. TIMES ARE COMPUTED FROM RESEARCHED SPEEDS. BALLPARK, NOT GOSPEL.
WHAT THE INTERNET SAYS
Not enough discussion about Gsm8k yet to call it. We found 37 posts but people did not say much either way.
34 POSTS · 3 COMMENTS · GITHUB · REDDIT
PEOPLE LOVE
- No fan mail yet. The crowd is still warming up.
PEOPLE COMPLAIN ABOUT
- No complaints yet. The internet is still looking for one. Suspicious.
LOOKING FOR AN ALTERNATIVE?
Other models the crowd has fully reviewed, starting with text models like this one.
DeepSeek V4 Flash> SOURCES (12)
- REDDIT · 2026-07-03 · 5 UPVOTESWhat's the best way to benchmark a model on SWE and gsm8k?
What's the best way to benchmark a model on SWE and gsm8k?
- REDDIT · 2025-05-13 · 4 UPVOTESGRPO Reward Decline After Convergence in Gemma-3-4B Fine-tuning
GRPO Reward Decline After Convergence in Gemma-3-4B Fine-tuning
- REDDIT · 2026-02-17 · 2 UPVOTESI Failed to Finetune a Model to Match a Character humour
I Failed to Finetune a Model to Match a Character humour
- GITHUB · 2026-09-03 · 2 UPVOTES[AMD][DSV4] Enable hicache on deepseek-v4 fp8 unified attn
[AMD][DSV4] Enable hicache on deepseek-v4 fp8 unified attn
- GITHUB · 2026-09-04 · 0 UPVOTES[NVBUG-6689820][test] unwaive Llama disaggregated serving test
[NVBUG-6689820][test] unwaive Llama disaggregated serving test
- GITHUB · 2026-09-02 · 0 UPVOTES[CI][ROCm] Stabilize Qwen 1.5 MoE GSM8K evals
[CI][ROCm] Stabilize Qwen 1.5 MoE GSM8K evals
- GITHUB · 2026-09-04 · 0 UPVOTES[CI][ROCm] Restore Wikitext coverage for Qwen OCP-MX
[CI][ROCm] Restore Wikitext coverage for Qwen OCP-MX
- GITHUB · 2026-09-04 · 0 UPVOTESDFlash2 acceptance regression on vLLM 0.28.0: 3.19 -> 2.66 tok/step, decode 120.5 -> 101.4 tok/s (prefill and GSM8K unchanged)
DFlash2 acceptance regression on vLLM 0.28.0: 3.19 -> 2.66 tok/step, decode 120.5 -> 101.4 tok/s (prefill and GSM8K unchanged)
- GITHUB · 2026-09-04 · 0 UPVOTESValidate archived MI355X MiniMax M3 vLLM disaggregation with srt-slurm
Validate archived MI355X MiniMax M3 vLLM disaggregation with srt-slurm
- GITHUB · 2026-09-04 · 0 UPVOTES[Model] Add DeepSeek-V4 CPU backend
[Model] Add DeepSeek-V4 CPU backend
- GITHUB · 2026-09-02 · 0 UPVOTESfeat(dflash2): co-train the DFlash2 drafter against vLLM rollout
feat(dflash2): co-train the DFlash2 drafter against vLLM rollout
- GITHUB · 2026-09-02 · 0 UPVOTES[ROCm] Fuse MLA dual RMSNorm + FP8 group quant for DeepSeek-R1
[ROCm] Fuse MLA dual RMSNorm + FP8 group quant for DeepSeek-R1