Capacity
What Spindle, the shared LoRA training server, can run right now, by model config. A config serves one base model at one maximum context length. A trainer is one GPU job for that config, and each trainer holds a fixed number of LoRA slots; one experiment takes one slot. A run on a free slot starts at once and adds no GPU cost while the trainer's other jobs keep it up. Pick a config to see its numbers, or open the experiments that fit it.
All configs
Free slots now
1381% of 16
Open LoRA slots on trainers that are already running. An experiment that takes one starts at once.
Trainers running
22% of 88 allowed
GPU jobs up across all configs. 0 starting, 86 more could start when a run needs one.
Configs with free compute
29% of 22
Model configs where an experiment can start without waiting.
Configs at capacity
00% of 22
Every allowed trainer is up and every slot is taken. A new run there waits for a slot to free.
Qwen2.5-7B-Instruct · 32k context
prod-qwen25-7b-instruct-lora-32k
Free now
Free slots now
788% of 8
Open LoRA slots on this config's 1 running trainer; 1 in use.
Trainers running
1of 4 allowed
3 more could start when a run needs one. Each trainer holds 8 LoRA slots.
Waiting
0models
Models queued for a slot on this config. A new run lines up behind them.
Released in the last hour
0slots
Slots freed by runs that finished in the last hour. A rough sign of how fast slots turn over.
Qwen2.5-3B · 32k context
prod-qwen25-3b-lora-32k
Free now
Free slots now
675% of 8
Open LoRA slots on this config's 1 running trainer; 2 in use.
Trainers running
1of 4 allowed
3 more could start when a run needs one. Each trainer holds 8 LoRA slots.
Waiting
0models
Models queued for a slot on this config. A new run lines up behind them.
Released in the last hour
0slots
Slots freed by runs that finished in the last hour. A rough sign of how fast slots turn over.
DeepSeek-V3.1 · 32k context
prod-deepseek-v31-lora-32k
After a trainer starts
Free slots now
0
No trainer is running for this config, so there are no open slots yet.
Trainers running
0of 4 allowed
4 more could start when a run needs one. Each trainer holds 2 LoRA slots.
Waiting
0models
Models queued for a slot on this config. A new run lines up behind them.
Released in the last hour
0slots
Slots freed by runs that finished in the last hour. A rough sign of how fast slots turn over.
DeepSeek-V4.1-Flash-1017 · 32k context
prod-deepseek-v41-flash-1017-lora-32k
After a trainer starts
Free slots now
0
No trainer is running for this config, so there are no open slots yet.
Trainers running
0of 4 allowed
4 more could start when a run needs one. Each trainer holds 2 LoRA slots.
Waiting
0models
Models queued for a slot on this config. A new run lines up behind them.
Released in the last hour
0slots
Slots freed by runs that finished in the last hour. A rough sign of how fast slots turn over.
DeepSeek-V4.1-Flash · 32k context
prod-deepseek-v41-flash-lora-32k
After a trainer starts
Free slots now
0
No trainer is running for this config, so there are no open slots yet.
Trainers running
0of 4 allowed
4 more could start when a run needs one. Each trainer holds 2 LoRA slots.
Waiting
0models
Models queued for a slot on this config. A new run lines up behind them.
Released in the last hour
0slots
Slots freed by runs that finished in the last hour. A rough sign of how fast slots turn over.
GLM-5.3 · 256k context
prod-glm53-lora-256k
After a trainer starts
Free slots now
0
No trainer is running for this config, so there are no open slots yet.
Trainers running
0of 4 allowed
4 more could start when a run needs one. Each trainer holds 4 LoRA slots.
Waiting
0models
Models queued for a slot on this config. A new run lines up behind them.
Released in the last hour
0slots
Slots freed by runs that finished in the last hour. A rough sign of how fast slots turn over.
gpt-oss-120b · 32k context
prod-gpt-oss-120b-lora-32k
After a trainer starts
Free slots now
0
No trainer is running for this config, so there are no open slots yet.
Trainers running
0of 4 allowed
4 more could start when a run needs one. Each trainer holds 6 LoRA slots.
Waiting
0models
Models queued for a slot on this config. A new run lines up behind them.
Released in the last hour
0slots
Slots freed by runs that finished in the last hour. A rough sign of how fast slots turn over.
gpt-oss-20b · 32k context
prod-gpt-oss-20b-lora-32k
After a trainer starts
Free slots now
0
No trainer is running for this config, so there are no open slots yet.
Trainers running
0of 4 allowed
4 more could start when a run needs one. Each trainer holds 8 LoRA slots.
Waiting
0models
Models queued for a slot on this config. A new run lines up behind them.
Released in the last hour
0slots
Slots freed by runs that finished in the last hour. A rough sign of how fast slots turn over.
Kimi-K2.6 · 32k context
prod-kimi-k26-lora-32k
After a trainer starts
Free slots now
0
No trainer is running for this config, so there are no open slots yet.
Trainers running
0of 4 allowed
4 more could start when a run needs one. Each trainer holds 2 LoRA slots.
Waiting
0models
Models queued for a slot on this config. A new run lines up behind them.
Released in the last hour
0slots
Slots freed by runs that finished in the last hour. A rough sign of how fast slots turn over.
NVIDIA-Nemotron-3-Nano-30B-A3B-BF16 · 64k context
prod-nemotron3-nano-30b-a3b-lora-64k
After a trainer starts
Free slots now
0
No trainer is running for this config, so there are no open slots yet.
Trainers running
0of 4 allowed
4 more could start when a run needs one. Each trainer holds 6 LoRA slots.
Waiting
0models
Models queued for a slot on this config. A new run lines up behind them.
Released in the last hour
0slots
Slots freed by runs that finished in the last hour. A rough sign of how fast slots turn over.
NVIDIA-Nemotron-3-Super-120B-A12B-BF16 · 64k context
prod-nemotron3-super-120b-a12b-lora-64k
After a trainer starts
Free slots now
0
No trainer is running for this config, so there are no open slots yet.
Trainers running
0of 4 allowed
4 more could start when a run needs one. Each trainer holds 4 LoRA slots.
Waiting
0models
Models queued for a slot on this config. A new run lines up behind them.
Released in the last hour
0slots
Slots freed by runs that finished in the last hour. A rough sign of how fast slots turn over.
NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16 · 64k context
prod-nemotron3-ultra-550b-a55b-lora-64k
After a trainer starts
Free slots now
0
No trainer is running for this config, so there are no open slots yet.
Trainers running
0of 4 allowed
4 more could start when a run needs one. Each trainer holds 4 LoRA slots.
Waiting
0models
Models queued for a slot on this config. A new run lines up behind them.
Released in the last hour
0slots
Slots freed by runs that finished in the last hour. A rough sign of how fast slots turn over.
NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16 · 64k context
prod-nemotron35-lightning-30b-a3b-lora-64k
After a trainer starts
Free slots now
0
No trainer is running for this config, so there are no open slots yet.
Trainers running
0of 4 allowed
4 more could start when a run needs one. Each trainer holds 4 LoRA slots.
Waiting
0models
Models queued for a slot on this config. A new run lines up behind them.
Released in the last hour
0slots
Slots freed by runs that finished in the last hour. A rough sign of how fast slots turn over.
Qwen3-4B · 32k context
prod-qwen3-4b-lora-32k
After a trainer starts
Free slots now
0
No trainer is running for this config, so there are no open slots yet.
Trainers running
0of 4 allowed
4 more could start when a run needs one. Each trainer holds 8 LoRA slots.
Waiting
0models
Models queued for a slot on this config. A new run lines up behind them.
Released in the last hour
1slot
Slots freed by runs that finished in the last hour. A rough sign of how fast slots turn over.
Qwen3-8B · 32k context
prod-qwen3-8b-lora-32k
After a trainer starts
Free slots now
0
No trainer is running for this config, so there are no open slots yet.
Trainers running
0of 4 allowed
4 more could start when a run needs one. Each trainer holds 8 LoRA slots.
Waiting
0models
Models queued for a slot on this config. A new run lines up behind them.
Released in the last hour
0slots
Slots freed by runs that finished in the last hour. A rough sign of how fast slots turn over.
Qwen3.5-2B · 32k context
prod-qwen35-2b-lora-32k
After a trainer starts
Free slots now
0
No trainer is running for this config, so there are no open slots yet.
Trainers running
0of 4 allowed
4 more could start when a run needs one. Each trainer holds 8 LoRA slots.
Waiting
0models
Models queued for a slot on this config. A new run lines up behind them.
Released in the last hour
0slots
Slots freed by runs that finished in the last hour. A rough sign of how fast slots turn over.
Qwen3.5-397B-A17B · 64k context
prod-qwen35-397b-a17b-lora-64k
After a trainer starts
Free slots now
0
No trainer is running for this config, so there are no open slots yet.
Trainers running
0of 4 allowed
4 more could start when a run needs one. Each trainer holds 2 LoRA slots.
Waiting
0models
Models queued for a slot on this config. A new run lines up behind them.
Released in the last hour
0slots
Slots freed by runs that finished in the last hour. A rough sign of how fast slots turn over.
Qwen3.5-4B · 32k context
prod-qwen35-4b-lora-32k
After a trainer starts
Free slots now
0
No trainer is running for this config, so there are no open slots yet.
Trainers running
0of 4 allowed
4 more could start when a run needs one. Each trainer holds 8 LoRA slots.
Waiting
0models
Models queued for a slot on this config. A new run lines up behind them.
Released in the last hour
0slots
Slots freed by runs that finished in the last hour. A rough sign of how fast slots turn over.
Qwen3.5-9B · 128k context
prod-qwen35-9b-lora-128k
After a trainer starts
Free slots now
0
No trainer is running for this config, so there are no open slots yet.
Trainers running
0of 4 allowed
4 more could start when a run needs one. Each trainer holds 8 LoRA slots.
Waiting
0models
Models queued for a slot on this config. A new run lines up behind them.
Released in the last hour
0slots
Slots freed by runs that finished in the last hour. A rough sign of how fast slots turn over.
Qwen3.5-9B-Base · 64k context
prod-qwen35-9b-lora-64k
After a trainer starts
Free slots now
0
No trainer is running for this config, so there are no open slots yet.
Trainers running
0of 4 allowed
4 more could start when a run needs one. Each trainer holds 6 LoRA slots.
Waiting
0models
Models queued for a slot on this config. A new run lines up behind them.
Released in the last hour
0slots
Slots freed by runs that finished in the last hour. A rough sign of how fast slots turn over.
Qwen3.6-35B-A3B · 32k context
prod-qwen36-35b-a3b-lora-32k
After a trainer starts
Free slots now
0
No trainer is running for this config, so there are no open slots yet.
Trainers running
0of 4 allowed
4 more could start when a run needs one. Each trainer holds 6 LoRA slots.
Waiting
0models
Models queued for a slot on this config. A new run lines up behind them.
Released in the last hour
0slots
Slots freed by runs that finished in the last hour. A rough sign of how fast slots turn over.
Qwen3.8-27B · 64k context
prod-qwen38-27b-lora-64k
After a trainer starts
Free slots now
0
No trainer is running for this config, so there are no open slots yet.
Trainers running
0of 4 allowed
4 more could start when a run needs one. Each trainer holds 6 LoRA slots.
Waiting
0models
Models queued for a slot on this config. A new run lines up behind them.
Released in the last hour
0slots
Slots freed by runs that finished in the last hour. A rough sign of how fast slots turn over.
| config | availability | trainers | slots per trainer | slots used | slots free | waiting | released, last hour | |
|---|---|---|---|---|---|---|---|---|
Qwen2.5-7B-Instruct · 32kprod-qwen25-7b-instruct-lora-32k
Free now |
Free now | 1 of 4 3 could start | 8 | 1 of 8 | 7 | 0 | 0 | |
Qwen2.5-3B · 32kprod-qwen25-3b-lora-32k
Free now |
Free now | 1 of 4 3 could start | 8 | 2 of 8 | 6 | 0 | 0 | |
DeepSeek-V3.1 · 32kprod-deepseek-v31-lora-32k
After a trainer starts |
After a trainer starts | 0 of 4 4 could start | 2 | 0 of 0 | 0 | 0 | 0 | |
DeepSeek-V4.1-Flash-1017 · 32kprod-deepseek-v41-flash-1017-lora-32k
After a trainer starts |
After a trainer starts | 0 of 4 4 could start | 2 | 0 of 0 | 0 | 0 | 0 | |
DeepSeek-V4.1-Flash · 32kprod-deepseek-v41-flash-lora-32k
After a trainer starts |
After a trainer starts | 0 of 4 4 could start | 2 | 0 of 0 | 0 | 0 | 0 | |
GLM-5.3 · 256kprod-glm53-lora-256k
After a trainer starts |
After a trainer starts | 0 of 4 4 could start | 4 | 0 of 0 | 0 | 0 | 0 | |
gpt-oss-120b · 32kprod-gpt-oss-120b-lora-32k
After a trainer starts |
After a trainer starts | 0 of 4 4 could start | 6 | 0 of 0 | 0 | 0 | 0 | |
gpt-oss-20b · 32kprod-gpt-oss-20b-lora-32k
After a trainer starts |
After a trainer starts | 0 of 4 4 could start | 8 | 0 of 0 | 0 | 0 | 0 | |
Kimi-K2.6 · 32kprod-kimi-k26-lora-32k
After a trainer starts |
After a trainer starts | 0 of 4 4 could start | 2 | 0 of 0 | 0 | 0 | 0 | |
NVIDIA-Nemotron-3-Nano-30B-A3B-BF16 · 64kprod-nemotron3-nano-30b-a3b-lora-64k
After a trainer starts |
After a trainer starts | 0 of 4 4 could start | 6 | 0 of 0 | 0 | 0 | 0 | |
NVIDIA-Nemotron-3-Super-120B-A12B-BF16 · 64kprod-nemotron3-super-120b-a12b-lora-64k
After a trainer starts |
After a trainer starts | 0 of 4 4 could start | 4 | 0 of 0 | 0 | 0 | 0 | |
NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16 · 64kprod-nemotron3-ultra-550b-a55b-lora-64k
After a trainer starts |
After a trainer starts | 0 of 4 4 could start | 4 | 0 of 0 | 0 | 0 | 0 | |
NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16 · 64kprod-nemotron35-lightning-30b-a3b-lora-64k
After a trainer starts |
After a trainer starts | 0 of 4 4 could start | 4 | 0 of 0 | 0 | 0 | 0 | |
Qwen3-4B · 32kprod-qwen3-4b-lora-32k
After a trainer starts |
After a trainer starts | 0 of 4 4 could start | 8 | 0 of 0 | 0 | 0 | 1 | |
Qwen3-8B · 32kprod-qwen3-8b-lora-32k
After a trainer starts |
After a trainer starts | 0 of 4 4 could start | 8 | 0 of 0 | 0 | 0 | 0 | |
Qwen3.5-2B · 32kprod-qwen35-2b-lora-32k
After a trainer starts |
After a trainer starts | 0 of 4 4 could start | 8 | 0 of 0 | 0 | 0 | 0 | |
Qwen3.5-397B-A17B · 64kprod-qwen35-397b-a17b-lora-64k
After a trainer starts |
After a trainer starts | 0 of 4 4 could start | 2 | 0 of 0 | 0 | 0 | 0 | |
Qwen3.5-4B · 32kprod-qwen35-4b-lora-32k
After a trainer starts |
After a trainer starts | 0 of 4 4 could start | 8 | 0 of 0 | 0 | 0 | 0 | |
Qwen3.5-9B · 128kprod-qwen35-9b-lora-128k
After a trainer starts |
After a trainer starts | 0 of 4 4 could start | 8 | 0 of 0 | 0 | 0 | 0 | |
Qwen3.5-9B-Base · 64kprod-qwen35-9b-lora-64k
After a trainer starts |
After a trainer starts | 0 of 4 4 could start | 6 | 0 of 0 | 0 | 0 | 0 | |
Qwen3.6-35B-A3B · 32kprod-qwen36-35b-a3b-lora-32k
After a trainer starts |
After a trainer starts | 0 of 4 4 could start | 6 | 0 of 0 | 0 | 0 | 0 | |
Qwen3.8-27B · 64kprod-qwen38-27b-lora-64k
After a trainer starts |
After a trainer starts | 0 of 4 4 could start | 6 | 0 of 0 | 0 | 0 | 0 |
- Config: one Spindle queue, named for its model and context length. Experiments on it share its trainers.
- Trainers: GPU jobs running the config now, out of the number Spindle may start. A trainer starts when a run needs a slot and none is free.
- Slots: LoRA slots per trainer, how many are in use, and how many are open on trainers already running.
- Waiting: models queued for a slot. Released, last hour: slots freed by runs that finished.