Dynamic TPU backend should support dynamic warpers and abort button

This commit is contained in:
Gnome Ann 2022-01-17 14:10:32 -05:00
parent 31735c4239
commit 3ba0e3f9d9
2 changed files with 23 additions and 22 deletions

View File

@ -1035,7 +1035,7 @@ else:
assert len(excluded_world_info) == len(generated) assert len(excluded_world_info) == len(generated)
regeneration_required = vars.lua_koboldbridge.regeneration_required regeneration_required = vars.lua_koboldbridge.regeneration_required
halt = not vars.lua_koboldbridge.generating or vars.generated_tkns >= vars.genamt halt = vars.abort or not vars.lua_koboldbridge.generating or vars.generated_tkns >= vars.genamt
vars.lua_koboldbridge.regeneration_required = False vars.lua_koboldbridge.regeneration_required = False
global past global past
@ -1061,6 +1061,15 @@ else:
def tpumtjgenerate_stopped_compiling_callback() -> None: def tpumtjgenerate_stopped_compiling_callback() -> None:
vars.compiling = False vars.compiling = False
def tpumtjgenerate_settings_callback() -> dict:
return {
"top_p": float(vars.top_p),
"temp": float(vars.temp),
"top_k": int(vars.top_k),
"tfs": float(vars.tfs),
"repetition_penalty": float(vars.rep_pen),
}
# If we're running Colab or OAI, we still need a tokenizer. # If we're running Colab or OAI, we still need a tokenizer.
if(vars.model == "Colab"): if(vars.model == "Colab"):
@ -3009,12 +3018,7 @@ def tpumtjgenerate(txt, minimum, maximum, found_entries=None):
tpu_mtj_backend.infer_dynamic, tpu_mtj_backend.infer_dynamic,
context, context,
gen_len = maximum-minimum+1, gen_len = maximum-minimum+1,
temp=vars.temp,
top_p=vars.top_p,
top_k=vars.top_k,
tfs=vars.tfs,
numseqs=vars.numseqs, numseqs=vars.numseqs,
repetition_penalty=vars.rep_pen,
soft_embeddings=vars.sp, soft_embeddings=vars.sp,
soft_tokens=soft_tokens, soft_tokens=soft_tokens,
excluded_world_info=found_entries, excluded_world_info=found_entries,
@ -3026,7 +3030,7 @@ def tpumtjgenerate(txt, minimum, maximum, found_entries=None):
assert vars.lua_koboldbridge.generated[r+1][c+1] is not None assert vars.lua_koboldbridge.generated[r+1][c+1] is not None
past[r, c] = vars.lua_koboldbridge.generated[r+1][c+1] past[r, c] = vars.lua_koboldbridge.generated[r+1][c+1]
if(halt or not regeneration_required): if(vars.abort or halt or not regeneration_required):
break break
print("(regeneration triggered)") print("(regeneration triggered)")

View File

@ -26,6 +26,15 @@ def warper_callback(logits) -> np.array:
def stopping_callback(generated, n_generated, excluded_world_info) -> Tuple[List[set], bool, bool]: def stopping_callback(generated, n_generated, excluded_world_info) -> Tuple[List[set], bool, bool]:
raise NotImplementedError("`tpu_mtj_backend.stopping_callback()` needs to be defined") raise NotImplementedError("`tpu_mtj_backend.stopping_callback()` needs to be defined")
def settings_callback() -> dict:
return {
"top_p": 0.9,
"temp": 0.5,
"top_k": 0,
"tfs": 1.0,
"repetition_penalty": 1.0,
}
def started_compiling_callback() -> None: def started_compiling_callback() -> None:
pass pass
@ -541,7 +550,7 @@ class PenalizingCausalTransformer(CausalTransformer):
out_axes=["shard", "batch", ...], out_axes=["shard", "batch", ...],
axis_resources={'shard': 'mp', 'batch': 'dp'}, axis_resources={'shard': 'mp', 'batch': 'dp'},
) )
def generate_dynamic(self, ctx, ctx_length, gen_length, numseqs, sampler_options, return_logits=False, soft_embeddings=None, excluded_world_info=None, use_callback=True): def generate_dynamic(self, ctx, ctx_length, gen_length, numseqs, return_logits=False, soft_embeddings=None, excluded_world_info=None, use_callback=True):
assert excluded_world_info is not None assert excluded_world_info is not None
assert not return_logits assert not return_logits
assert gen_length.ndim == 1 assert gen_length.ndim == 1
@ -560,7 +569,6 @@ class PenalizingCausalTransformer(CausalTransformer):
] ]
for i in range(numseqs) for i in range(numseqs)
] ]
repetition_penalty = sampler_options.pop("repetition_penalty", 1.0)
n_generated = 0 n_generated = 0
regeneration_required = False regeneration_required = False
halt = False halt = False
@ -576,6 +584,8 @@ class PenalizingCausalTransformer(CausalTransformer):
logits = warper_callback(logits) logits = warper_callback(logits)
for i in range(numseqs): for i in range(numseqs):
sample_data[i][2] = logits[i] sample_data[i][2] = logits[i]
sampler_options = settings_callback()
repetition_penalty = sampler_options.pop("repetition_penalty", 1.0)
sample_data, sample_key = sample_func(sample_data, sample_key, _numseqs_aux, badwords, repetition_penalty, sampler_options) sample_data, sample_key = sample_func(sample_data, sample_key, _numseqs_aux, badwords, repetition_penalty, sampler_options)
n_generated += 1 n_generated += 1
for i in range(numseqs): for i in range(numseqs):
@ -611,11 +621,6 @@ class PenalizingCausalTransformer(CausalTransformer):
def infer_dynamic( def infer_dynamic(
context: np.array, context: np.array,
top_p=0.9,
temp=0.5,
top_k=0,
tfs=1.0,
repetition_penalty=1.0,
numseqs=1, numseqs=1,
gen_len=80, gen_len=80,
soft_embeddings: Optional[np.array] = None, soft_embeddings: Optional[np.array] = None,
@ -634,19 +639,11 @@ def infer_dynamic(
padded_tokens = np.pad(tokens, ((0, 0), (pad_amount, 0)), constant_values=pad_token_id) padded_tokens = np.pad(tokens, ((0, 0), (pad_amount, 0)), constant_values=pad_token_id)
batched_tokens = np.array([padded_tokens] * total_batch) batched_tokens = np.array([padded_tokens] * total_batch)
samples = [] samples = []
generator_params = {
"temp": float(temp),
"top_p": float(top_p),
"tfs": float(tfs),
"repetition_penalty": float(repetition_penalty),
"top_k": int(top_k),
}
output = network.generate_dynamic( output = network.generate_dynamic(
batched_tokens, batched_tokens,
np.ones(total_batch, dtype=np.uint32) * provided_ctx, np.ones(total_batch, dtype=np.uint32) * provided_ctx,
np.ones(total_batch, dtype=np.uint32) * gen_len, np.ones(total_batch, dtype=np.uint32) * gen_len,
numseqs, numseqs,
generator_params,
soft_embeddings=soft_embeddings, soft_embeddings=soft_embeddings,
excluded_world_info=excluded_world_info, excluded_world_info=excluded_world_info,
use_callback=use_callback, use_callback=use_callback,