Skip to content

added parameter for KV cache tokens contribution to LLM impacts calcu… #200

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
gregoirevilde-se wants to merge 1 commit into
base: main
Choose a base branch
from
Open

added parameter for KV cache tokens contribution to LLM impacts calcu… #200

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
16 changes: 13 additions & 3 deletions ecologits/impacts/llm.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -41,6 +41,7 @@ def gpu_energy(
gpu_energy_alpha: float,
gpu_energy_beta: float,
gpu_energy_gamma: float,
cache_token_count: float = 0,
) -> ValueOrRange:
"""
Compute energy consumption of a single GPU.
Expand All @@ -52,14 +53,15 @@ def gpu_energy(
gpu_energy_alpha: Alpha coefficient of the energy regression.
gpu_energy_beta: Beta coefficient of the energy regression.
gpu_energy_gamma: Beta coefficient of the energy regression.
cache_toke_count: Number of tokens read from KV cache.

Returns:
The energy consumption of a single GPU in kWh.
"""
gpu_energy_per_token = gpu_energy_alpha * math.exp(gpu_energy_beta * batch_size) * model_active_parameter_count + \
gpu_energy_gamma
gpu_energy_per_token /= 1000 # convert to kWh
return output_token_count * gpu_energy_per_token
return (output_token_count + cache_token_count / 6) * gpu_energy_per_token


@dag.asset
Expand All @@ -70,7 +72,8 @@ def generation_latency(
latency_alpha: float,
latency_beta: float,
latency_gamma: float,
request_latency: float
request_latency: float,
cache_token_count: float = 0,
) -> ValueOrRange:
"""
Compute the token generation latency in seconds.
Expand All @@ -82,12 +85,13 @@ def generation_latency(
latency_alpha: Alpha coefficient of the latency regression.
latency_beta: Beta coefficient of the latency regression.
latency_gamma: Gamma coefficient of the latency regression.
cache_toke_count: Number of tokens read from KV cache.

Returns:
The token generation latency in seconds.
"""
latency_per_token = latency_alpha * model_active_parameter_count + latency_beta * batch_size + latency_gamma
gpu_latency = output_token_count * latency_per_token
gpu_latency = (output_token_count + cache_token_count / 6) * latency_per_token
if request_latency < gpu_latency:
return request_latency
return gpu_latency
Expand Down Expand Up @@ -394,6 +398,7 @@ def compute_llm_impacts_dag(
if_electricity_mix_wue: float,
datacenter_pue: ValueOrRange,
datacenter_wue: ValueOrRange,
cache_token_count: float = 0,
model_quantization_bits: Optional[int] = MODEL_QUANTIZATION_BITS,
gpu_energy_alpha: Optional[float] = GPU_ENERGY_ALPHA,
gpu_energy_beta: Optional[float] = GPU_ENERGY_BETA,
Expand Down Expand Up @@ -427,6 +432,7 @@ def compute_llm_impacts_dag(
if_electricity_mix_wue: WCF impact factor of electricity consumption in L / kWh.
datacenter_wue: Water Usage Effectiveness of the data center in L/kWh.
datacenter_pue: Power Usage Effectiveness of the data center.
cache_toke_count: Number of tokens read from KV cache.
model_quantization_bits: Number of bits used to represent the model weights.
gpu_energy_alpha: Alpha coefficient of the "GPU energy" regression.
gpu_energy_beta: Beta coefficient of the "GPU energy" regression.
Expand All @@ -453,6 +459,7 @@ def compute_llm_impacts_dag(
model_total_parameter_count=model_total_parameter_count,
model_quantization_bits=model_quantization_bits,
output_token_count=output_token_count,
cache_token_count=cache_token_count,
request_latency=request_latency,
if_electricity_mix_gwp=if_electricity_mix_gwp,
if_electricity_mix_adpe=if_electricity_mix_adpe,
Expand Down Expand Up @@ -491,6 +498,7 @@ def compute_llm_impacts(
if_electricity_mix_wue: float,
datacenter_pue: ValueOrRange,
datacenter_wue: ValueOrRange,
cache_token_count: float = 0,
request_latency: Optional[float] = None,
**kwargs: Any
) -> Impacts:
Expand All @@ -507,6 +515,7 @@ def compute_llm_impacts(
if_electricity_mix_wue: WCF impact factor of electricity consumption in L / kWh.
datacenter_wue: Water Usage Effectiveness of the data center in L/kWh.
datacenter_pue: Power Usage Effectiveness of the data center.
cache_toke_count: Number of tokens read from KV cache.
request_latency: Measured request latency in seconds.
**kwargs: Any other optional parameter.
Returns:
Expand Down Expand Up @@ -543,6 +552,7 @@ def compute_llm_impacts(
if_electricity_mix_wue=if_electricity_mix_wue,
datacenter_pue=datacenter_pue,
datacenter_wue=datacenter_wue,
cache_token_count=cache_token_count,
**kwargs
)
for field in fields:
Expand Down