fix batch count with > 1 slave requesting duplicate seeds

pull/15/head
unknown 2023-06-02 23:26:31 -05:00
parent 7ac7541b98
commit a94433936f
No known key found for this signature in database
GPG Key ID: CA376082283AF69A
2 changed files with 50 additions and 20 deletions

View File

@ -167,7 +167,7 @@ class Script(scripts.Script):
def add_to_gallery(processed, p):
"""adds generated images to the image gallery after waiting for all workers to finish"""
def processed_inject_image(image, info_index, save_path_override=None):
def processed_inject_image(image, info_index, iteration: int, save_path_override=None, grid=False):
image_params: json = worker.response["parameters"]
image_info_post: json = json.loads(worker.response["info"]) # image info known after processing
@ -179,35 +179,45 @@ class Script(scripts.Script):
except Exception:
# like with controlnet masks, there isn't always full post-gen info, so we use the first images'
logger.debug(f"Image at index {i} for '{worker.uuid}' was missing some post-generation data")
processed_inject_image(image=image, info_index=0)
processed_inject_image(image=image, info_index=0, iteration=iteration)
return
processed.all_prompts.append(image_params["prompt"])
processed.images.append(image) # actual received image
# generate info-text string
images_per_batch = p.n_iter * p.batch_size
# zero-indexed position of image in total batch (so including master results)
true_image_pos = len(processed.images) - 1
num_remote_images = images_per_batch * p.batch_size
if p.n_iter > 1: # if splitting by batch count
num_remote_images *= p.n_iter - 1
total_images = num_remote_images + images_per_batch
info_text_used_seed_index = info_index + p.n_iter * p.batch_size
if iteration != 0:
logger.debug(f"iteration {iteration}/{p.n_iter}, image {true_image_pos + 1}/{total_images}, info-index: {info_index}, used seed index {info_text_used_seed_index}")
info_text = processing.create_infotext(
p,
processed.all_prompts,
processed.all_seeds,
processed.all_subseeds,
p=p,
all_prompts=processed.all_prompts,
all_seeds=processed.all_seeds,
all_subseeds=processed.all_subseeds,
# comments=[""], # unimplemented upstream :(
# we don't need the "true_image_pos" like below with save_image because this method does it for us
position_in_batch=info_index, # zero-indexed
iteration=p.n_iter # if batch count is 2 p.n_iter will be 2
position_in_batch=true_image_pos,
iteration=0 # if batch count is 2 p.n_iter will be 2
)
processed.infotexts.append(info_text)
# zero-indexed position of image in total batch (so including master results)
true_image_pos = info_index + (p.n_iter * p.batch_size)
# automatically save received image to local disk if desired
if cmd_opts.distributed_remotes_autosave:
save_image(
image=image,
path=p.outpath_samples if save_path_override is None else save_path_override,
basename="",
seed=processed.all_seeds[true_image_pos],
prompt=processed.all_prompts[true_image_pos],
seed=processed.all_seeds[-1],
prompt=processed.all_prompts[-1],
info=info_text,
extension=opts.samples_format
)
@ -220,6 +230,7 @@ class Script(scripts.Script):
for thread in Script.worker_threads:
thread.join()
spoofed_iteration = p.n_iter
for worker in Script.world.workers:
# if it fails here then that means that the response_cache global var is not being filled for some reason
expected_images = 1
@ -237,17 +248,25 @@ class Script(scripts.Script):
logger.warning(f"Worker '{worker.uuid}' had nothing")
continue
injected_to_iteration = 0
images_per_iteration = Script.world.get_current_output_size()
# visibly add work from workers to the image gallery
for i in range(0, len(images)):
image_bytes = base64.b64decode(images[i])
image = Image.open(io.BytesIO(image_bytes))
# inject image
processed_inject_image(image=image, info_index=i)
processed_inject_image(image=image, info_index=i, iteration=spoofed_iteration)
if injected_to_iteration >= images_per_iteration - 1:
spoofed_iteration += 1
injected_to_iteration = 0
else:
injected_to_iteration += 1
# generate and inject grid
grid = processing.images.image_grid(processed.images, len(processed.images))
processed_inject_image(image=grid, info_index=0, save_path_override=p.outpath_grids)
# grid = processing.images.image_grid(processed.images, len(processed.images))
# processed_inject_image(image=grid, info_index=0, save_path_override=p.outpath_grids, iteration=0, grid=True)
p.batch_size = len(processed.images)
"""
@ -347,23 +366,34 @@ class Script(scripts.Script):
# start generating images assigned to remote machines
sync = False # should only really to sync once per job
Script.world.optimize_jobs(payload) # optimize work assignment before dispatching
started_jobs = []
for job in Script.world.jobs:
payload_temp = copy.deepcopy(payload)
if job.worker.master:
started_jobs.append(job)
if job.batch_size < 1 or job.worker.master:
continue
payload['batch_size'] = job.batch_size
payload['subseed'] += 1 * p.n_iter
payload['seed'] += (1 * p.n_iter) if payload['subseed_strength'] == 0 else 0
prior_images = 0
for j in started_jobs:
prior_images += j.batch_size * p.n_iter
payload_temp['batch_size'] = job.batch_size
payload_temp['subseed'] += prior_images
payload_temp['seed'] += prior_images if payload_temp['subseed_strength'] == 0 else 0
logger.debug(f"{job.worker.uuid}' job given starting seed is {payload_temp['seed']} with {prior_images} coming before it")
if job.worker.loaded_model != name or job.worker.loaded_vae != vae:
sync = True
job.worker.loaded_model = name
job.worker.loaded_vae = vae
t = Thread(target=job.worker.request, args=(payload, option_payload, sync, ))
t = Thread(target=job.worker.request, args=(payload_temp, option_payload, sync, ))
t.start()
Script.worker_threads.append(t)
started_jobs.append(job)
# if master batch size was changed again due to optimization change it to the updated value
p.batch_size = Script.world.get_master_batch_size()

View File

@ -296,7 +296,7 @@ class Worker:
if self.benchmarked:
eta = self.batch_eta(payload=payload)
logger.info(f"worker '{self.uuid}' predicts it will take {eta:.3f}s to generate {payload['batch_size']} image("
logger.debug(f"worker '{self.uuid}' predicts it will take {eta:.3f}s to generate {payload['batch_size']} image("
f"s) at a speed of {self.avg_ipm} ipm\n")
try: