Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 79282.72345428595,
-      "required_pages_per_second_upper_estimate": 0.005484962769830832
+      "required_token_positions_per_second": 78849.39479656596,
+      "required_pages_per_second_upper_estimate": 0.005454984087828752
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9910.340431785744,
-      "required_pages_per_second_upper_estimate": 0.004149252587347242
+      "required_token_positions_per_second": 9856.174349570745,
+      "required_pages_per_second_upper_estimate": 0.004126574379840243
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19820.680863571488,
-      "required_pages_per_second_upper_estimate": 0.004128125117377078
+      "required_token_positions_per_second": 19712.34869914149,
+      "required_pages_per_second_upper_estimate": 0.004105562384437605
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 29731.02129535723,
-      "required_pages_per_second_upper_estimate": 0.0017823293300156986
+      "required_token_positions_per_second": 29568.523048712235,
+      "required_pages_per_second_upper_estimate": 0.0017725878082497874
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 29731.02129535723,
-      "required_pages_per_second_upper_estimate": 0.002311456132771857
+      "required_token_positions_per_second": 29568.523048712235,
+      "required_pages_per_second_upper_estimate": 0.002298822608849458
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 24,
-      "required_token_positions_per_second": 29731.02129535723,
-      "required_pages_per_second_upper_estimate": 0.38135761849331373
+      "required_token_positions_per_second": 29568.523048712235,
+      "required_pages_per_second_upper_estimate": 0.3792732654623752
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,27 +158,40 @@
     "prefetch_batches": 65,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 1.304301914000007,
-    "producer_p95_seconds": 2.591754088000016,
-    "gpu_update_seconds": 0.9299377820000245,
-    "producer_batches_per_second": 0.7666936537210316,
-    "gpu_batches_per_second": 1.0753407586573074,
-    "producer_headroom": 0.7129773958148309,
-    "sustainable": false,
-    "buffer_seconds": 60.445955830001594,
-    "buffer_drain_seconds": 210.5964998875337,
+    "producer_batch_seconds": 0.04021698036923453,
+    "producer_p95_seconds": 0.1268318909999948,
+    "gpu_update_seconds": 0.9350483943500222,
+    "producer_batches_per_second": 24.86511893282239,
+    "gpu_batches_per_second": 1.0694633625836312,
+    "producer_headroom": 23.250089533457913,
+    "sustainable": true,
+    "buffer_seconds": 60.77814563275144,
+    "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 7,
-    "max_pending_pages": 1769,
-    "generation_bytes_per_second": 0,
-    "upload_bytes_per_second": null,
-    "upload_sustainable": null,
+    "upload_pages_per_commit": 23,
+    "max_pending_pages": 4096,
+    "generation_bytes_per_second": 412574.99951053865,
+    "upload_bytes_per_second": 85005.02186575368,
+    "upload_sustainable": false,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 0.0,
+    "estimated_full_shards_per_hour": 88.52899064051742,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
-  "upload_measurements": [],
+  "upload_measurements": [
+    {
+      "seconds": 1.386034201999962,
+      "bytes": 47132,
+      "pages": 1,
+      "shards": 1
+    },
+    {
+      "seconds": 1.0035069749999934,
+      "bytes": 155991,
+      "pages": 4,
+      "shards": 1
+    }
+  ],
   "cache_key": {
     "version": 1,
     "kind": "data",

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -0,0 +1 @@
+{"step": 604500, "loss": 0.636208713054657, "lr": 0.0001, "grad_norm": 0.16678688842682265, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88481063206, "tokens_per_second": 152754.29312491545, "data_wait_s": 1.4469000007011346e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:46edb1e790e35d35a38b1a8f61a7ab82bddc52cd2c76e1fb71e54972c526746d
-size 165
+oid sha256:7e77d6fa6e8786044746245471be3e87e95daa5bfbd30666f2bcbeace3ec6981
+size 2283