Bump vLLM and TRL

[lewtun]

I've tested the SFT and GRPO sample scripts still run :)

AIME24 eval for Llama 8B also fine:

Task	Version	Metric	Value		Stderr
all		math_pass@1:1_samples	0.4000	±	0.0910
		math_pass@1:4_samples	0.4667	±	0.0696
		math_pass@1:8_samples	0.4667	±	0.0656
		math_pass@1:16_samples	0.4604	±	0.0639
		math_pass@1:32_samples	0.4594	±	0.0634
		math_pass@1:64_samples	0.4510	±	0.0640
lighteval:aime24:0	2	math_pass@1:1_samples	0.4000	±	0.0910
		math_pass@1:4_samples	0.4667	±	0.0696
		math_pass@1:8_samples	0.4667	±	0.0656
		math_pass@1:16_samples	0.4604	±	0.0639
		math_pass@1:32_samples	0.4594	±	0.0634
		math_pass@1:64_samples	0.4510	±	0.0640

Ran with

export VLLM_WORKER_MULTIPROC_METHOD=spawn
NUM_GPUS=8
MODEL=deepseek-ai/DeepSeek-R1-Distill-Llama-8B
MODEL_ARGS="model_name=$MODEL,dtype=bfloat16,data_parallel_size=$NUM_GPUS,max_model_length=32768,gpu_memory_utilization=0.8,generation_parameters={max_new_tokens:32768,temperature:0.6,top_p:0.95}"
TASK=aime24
OUTPUT_DIR=data/evals/$MODEL

lighteval vllm $MODEL_ARGS "lighteval|$TASK|0|0" \
    --use-chat-template \
    --output-dir $OUTPUT_DIR