You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
We've made some changes to the platform that will impact start times for Sessions and Runs.
9
+
10
+
As a result of these changes, you'll experience longer start times for Sessions and Runs that use the `p3.2xlarge` instance type. If you're looking for a faster start time, we suggest using the `g4dn.xlarge` instance type instead.
11
+
12
+
**In future Grid releases, the following instance types will be supported:**
13
+
14
+
| Name | CPU | GPU | Memory | Accelerator | numberOfAccelerators acceleratorType availableMemory |
15
+
| :--- | :--- | :--- | :--- | :--- | :--- |
16
+
|**m5a.large (recommended for fast startup times)**| 2 | 0 | 8 | CPU | 2_CPU_8GB |
17
+
| m5a.2xlarge | 8 | 0 | 32 | CPU | 8_CPU_32GB |
18
+
|**g4dn.xlarge (recommended for fast startup times)**| 4 | 1 | 16 | T4 | 1_T4_16GB |
We closely monitor usage of Grid and are always looking for improvements that will make the platform more straightforward, easier to use, and cost-effective.
27
+
In changing how we manage certain instance types, we're able to offer faster start times on cheaper instances. Managing these instance types is a key area that will make Grid more sustainable and less expensive to use in the long term. We always want to ensure that Grid users are getting the compute resources they need at a price that is fair and transparent.
28
+
29
+
### BYOC Instance Types
30
+
31
+
If you are currently using the BYOC feature, you will continue to have access to the full list of [supported AWS instance types](../docs/platform/3_machines.md#machines). If you are not currently using BYOC and want access to or information about additional instance types, reach out to us at support@grid.ai.
32
+
33
+
34
+
If you've got questions about these changes, reach out to us at support@grid.ai.
35
+
36
+
## Fixes and Enhancements
37
+
38
+
- Adds UI support for [skipping parameter evaluation](../docs/features/runs/1_Creating%20Runs/1_Basic%20Runs/3_sweep-syntax.md#skipping-parameter-evaluation) when running hyperparemeter sweeps
39
+
40
+
- Improvements to the process of integrating Grid with public and private Github organizations
41
+
42
+
- BYOC users: Fixes issue with starting runs with unavailable instance types. If the default instance type is not available, the first instance in the specified list of instances will be used instead
43
+
44
+
- Stability improvements in the UI to make analzying experiment results a better experience
45
+
46
+
- Better error messaging in the CLI
47
+
48
+
- Fixes CLI issue where users could only retrieve the 50 most recent runs. To request details for a specific run in your run history, use `grid status RUN_NAME`
49
+
50
+
## :warning: Known Issues
51
+
52
+
- When creating a run in the UI, specify the path to the Github repo where the script is located. Providing the URL to the specific script is not currently supported.
53
+
54
+
- When creating a Datastore, data directories that contain soft symlinks files will cause the Datastore upload to fail. To prevent this failure, update soft symlinks to hard links.
Copy file name to clipboardexpand all lines: docs/changelog.md
+56
Original file line number
Diff line number
Diff line change
@@ -12,6 +12,62 @@ Upgrade your CLI with `pip install lightning-grid --upgrade`
12
12
:heart: Find us in our [Slack Community](http://gridai-community.slack.com) to say hi and/or to express your thoughts/questions.
13
13
14
14
---
15
+
## :zap: June 7, 2022
16
+
17
+
**CLI version: 0.8.58**
18
+
19
+
Today's release includes several bug fixes that improve the Grid experience.
20
+
21
+
## Grid Cloud Instance Types
22
+
23
+
We've made some changes to the platform that will impact start times for Sessions and Runs.
24
+
25
+
As a result of these changes, you'll experience longer start times for Sessions and Runs that use the `p3.2xlarge` instance type. If you're looking for a faster start time, we suggest using the `g4dn.xlarge instance type instead`.
26
+
27
+
**In future Grid releases, the following instance types will be supported:**
28
+
29
+
| Name | CPU | GPU | Memory | Accelerator | numberOfAccelerators acceleratorType availableMemory |
30
+
| :--- | :--- | :--- | :--- | :--- | :--- |
31
+
| m5a.large (recommended for fast startup times) | 2 | 0 | 8 | CPU | 2_CPU_8GB |
32
+
| m5a.2xlarge | 8 | 0 | 32 | CPU | 8_CPU_32GB |
33
+
| g4dn.xlarge (recommended for fast startup times) | 4 | 1 | 16 | T4 | 1_T4_16GB |
We closely monitor usage of Grid and are always looking for improvements that will make the platform more straightforward, easier to use, and cost-effective.
42
+
In changing how we manage certain instance types, we're able to offer faster start times on cheaper instances. Managing these instance types is a key area that will make Grid more sustainable and less expensive to use in the long term. We always want to ensure that Grid users are getting the compute resources they need at a price that is fair and transparent.
43
+
44
+
### BYOC Instance Types
45
+
46
+
If you are currently using the BYOC feature, you will continue to have access to the full list of [supported AWS instance types](../docs/platform/3_machines.md#machines). If you are not currently using BYOC and want access to or information about additional instance types, reach out to us at support@grid.ai.
47
+
48
+
49
+
If you've got questions about these changes, reach out to us at support@grid.ai.
50
+
51
+
## Fixes and Enhancements
52
+
53
+
- Improvements to the process of integrating Grid with public and private Github organizations
54
+
55
+
- BYOC users: Fixes issue with starting runs with unavailable instance types. If the default instance type is not available, the first instance in the specified list of instances will be used instead.
56
+
57
+
- Better error messaging in the CLI!
58
+
59
+
- Fixes CLI issue where users could only retrieve the 50 most recent runs. To request details for a specific run in your run history, use `grid status RUN_NAME`.
60
+
61
+
## :warning: Known Issues
62
+
63
+
- When creating a run in the UI, specify the path to the Github repo where the script is located. Providing the URL to the specific script is not currently supported.
64
+
65
+
- When creating a Datastore, data directories that contain soft symlinks files will cause the Datastore upload to fail. To prevent this failure, update soft symlinks to hard links.
Copy file name to clipboardexpand all lines: docs/cli.md
+2-2
Original file line number
Diff line number
Diff line change
@@ -481,8 +481,8 @@ grid run [OPTIONS] [RUN_COMMAND]...
481
481
|`--seed`| text | Seed value for the `random_search` strategy | None |
482
482
|`--instance_type`| text | Instance type to start training session in |`t2.medium`|
483
483
|`--gpus`| integer | Number of GPUs to allocate per experiment |`0`|
484
-
|`--cpus`| integer | Number of CPUs to allocate per experiment |`1`|
485
-
|`--memory`| text | How much memory an experiment needs |`100`|
484
+
|`--cpus`| integer | Number of CPUs to allocate per experiment. This parameter also affects memory (RAM) allocating for your experiment using the following rule: the amount of memory for the experiments will be allocated in the same proportion as the CPU allocated for the instance type chosen for the experiments. For example, if you plan to choose a machine with 16 CPUs and 64 Gb RAM and use a default value of CPUs (1 CPU) for your experiments, 1/16 * 64 Gb = 4 Gb of RAM will be allocated per each experiment.|`1`|
485
+
|`--memory`| text | How much disk memory (storage) an experiment needs, Gb|`100`|
486
486
|`--datastore_name`| text | Datastore name to be mounted in training | None |
487
487
|`--datastore_version`| integer | Datastore version to be mounted in training | None |
488
488
|`--datastore_mount_dir`| text | Directory to mount Datastore in training job. The default datastore mount location is /datastores | None |
0 commit comments