[bloom inference scripts] improvements

scripts/bloom-inference-scripts/README.md

@@ -18,8 +18,9 @@ Throughput in msecs on 8x80GB gpus:
 | accelerate   int8 | 286.56 | 40.92 | 22.65 | 13.27 |  oom |      |      |      |
 | ds-inference fp16 |  44.02 |  5.70 |  3.01 |  1.68 | 1.00 | 0.69 |  oom |      |
 | ds-inference int8 |  89.09 | 11.44 |  5.88 |  3.09 | 1.71 | 1.02 | 0.71 | oom  |
-| ds-zero           |    283 | 34.88 |   oom |       |      |      |      |      |
-|                   |        |       |       |       |      |      |      |      |
+| ds-zero      bf16 |    283 | 34.88 |   oom |       |      |      |      |      |
+
+note: Since Deepspeed-ZeRO can process multiple generate streams in parallel its throughput can be further divided by 8 or 16, depending on whether 8 or 16 gpus were used during the generate. and, of course, it means that it can process a bs of 64 in the case of 8x80 A100 (the table above).
 
 Start to ready to generate in secs (mainly loading and data preparation time):
 
@@ -39,7 +40,6 @@ Throughput in msecs 4x80GB A100:
 | :---------------- | :----- | :---- | :---- | :---- | :--- | :--- |
 | accelerate   int8 | 284.15 | 40.14 | 21.97 |  oom  |      |      |
 | ds-inference int8 | 156.51 | 20.11 | 10.38 |  5.50 | 2.96 | oom  |
-|                   |        |       |       |       |      |      |
 
 To get the benchmark results simply add `--benchmark` to any of these 3 scripts discussed below.
 
@@ -145,6 +145,8 @@ Note that the script currently runs the same inputs on all GPUs, but you can run
 deepspeed --num_gpus 8 scripts/bloom-inference-scripts/bloom-ds-zero-inference.py --name bigscience/bloom --batch_size 1 --benchmark 2>&1 | tee bloom-ds-zero-inference_bs=1.txt
 ```
 
+Please remember that with ZeRO the user can generate multiple unique streams at the same time - and thus the overall performance should be throughput in secs/token divided by number of participating gpus - so 8x to 16x faster depending on whether 8 or 16 gpus were used!
+
 You can also try the offloading solutions with just one small GPU, which will take a long time to run, but if you don't have 8 huge GPUs this is as good as it gets.
 
 

scripts/bloom-inference-scripts/bloom-accelerate-inference.py

@@ -163,26 +163,22 @@ def generate():
 
     return zip(inputs, outputs, total_new_tokens)
 
-# warmup is a must if measuring speed as it's when all the optimizations are performed
-# e.g. on 8x80 a100 the first pass of 100 tokens takes 23sec, and the next one is 4secs
-_ = generate()
-
+print_rank0(f"*** Running generate")
 t_generate_start = time.time()
 generated = generate()
 t_generate_span = time.time() - t_generate_start
 for i,o,_ in generated:
     print_rank0(f"{'-'*60}\nin={i}\nout={o}\n")
 
 
+### Benchmark
+
 if args.benchmark:
+    # clear cache / free memory
     torch.cuda.empty_cache()
     gc.collect()
 
-### Benchmark
-
-if args.benchmark:
     print_rank0(f"*** Running benchmark")
-
     # warm up
     for i in range(1):
         _ = generate()

scripts/bloom-inference-scripts/bloom-ds-inference.py

@@ -251,8 +251,10 @@ def generate():
 
 # warmup is a must if measuring speed as it's when all the optimizations are performed
 # e.g. on 8x80 a100 the first pass of 100 tokens takes 23sec, and the next one is 4secs
+print_rank0(f"*** Running generate warmup")
 _ = generate()
 
+print_rank0(f"*** Running generate")
 t_generate_start = time.time()
 generated = generate()
 t_generate_span = time.time() - t_generate_start

scripts/bloom-inference-scripts/bloom-ds-zero-inference.py

@@ -105,7 +105,7 @@ def print_rank0(*msg):
         device="nvme",
         pin_memory=True,
         nvme_path=args.nvme_offload_path,
-        buffer_size=6e8
+        buffer_size=4e9,
     )
 
 dschf = HfDeepSpeedConfig(ds_config) # this tells from_pretrained to instantiate directly on gpus
@@ -130,6 +130,7 @@ def print_rank0(*msg):
 
 if args.benchmark:
     t_ready = time.time()
+    deepspeed.runtime.utils.see_memory_usage('start-of-generate', force=True)
 
 
 ### Generate
@@ -175,25 +176,22 @@ def generate():
 
 # XXX: this is currently doing world_size streams on world_size gpus, so we can feed it different inputs on each! and hence the time can be divided by world_size
 
-# warmup is a must if measuring speed as it's when all the optimizations are performed
-# e.g. on 8x80 a100 the first pass of 100 tokens takes 23sec, and the next one is 4secs
-_ = generate()
-
+print_rank0(f"*** Running generate")
 t_generate_start = time.time()
 pairs = generate()
 t_generate_span = time.time() - t_generate_start
 for i,o,_ in pairs:
     print_rank0(f"{'-'*60}\nin={i}\nout={o}\n")
 
 
+### Benchmark
+
 if args.benchmark:
+    # clear cache / free memory
     torch.cuda.empty_cache()
     gc.collect()
-    deepspeed.runtime.utils.see_memory_usage('end-of-run', force=True)
+    deepspeed.runtime.utils.see_memory_usage('end-of-generate', force=True)
 
-### Benchmark
-
-if args.benchmark:
     print_rank0(f"*** Running benchmark")
 
     # warm up