Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76458.36287815156,
-      "required_pages_per_second_upper_estimate": 0.006627217112329103
+      "required_token_positions_per_second": 76283.20649797843,
+      "required_pages_per_second_upper_estimate": 0.006612035001225484
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9557.295359768945,
-      "required_pages_per_second_upper_estimate": 0.003812548013311371
+      "required_token_positions_per_second": 9535.400812247304,
+      "required_pages_per_second_upper_estimate": 0.0038038139509523313
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19114.59071953789,
-      "required_pages_per_second_upper_estimate": 0.0044218241576580945
+      "required_token_positions_per_second": 19070.80162449461,
+      "required_pages_per_second_upper_estimate": 0.004411694321181583
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28671.88607930683,
-      "required_pages_per_second_upper_estimate": 0.0017459674324693524
+      "required_token_positions_per_second": 28606.202436741914,
+      "required_pages_per_second_upper_estimate": 0.0017419676432525882
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28671.88607930683,
-      "required_pages_per_second_upper_estimate": 0.0022072411592455403
+      "required_token_positions_per_second": 28606.202436741914,
+      "required_pages_per_second_upper_estimate": 0.0022021846506169334
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28671.88607930683,
-      "required_pages_per_second_upper_estimate": 0.35600886647512114
+      "required_token_positions_per_second": 28606.202436741914,
+      "required_pages_per_second_upper_estimate": 0.35519329546347533
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,17 +158,17 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.05567971638284064,
-    "producer_p95_seconds": 0.15143594999972265,
-    "gpu_update_seconds": 0.964289545638025,
-    "producer_batches_per_second": 17.959861597071278,
-    "gpu_batches_per_second": 1.0370329166415955,
-    "producer_headroom": 17.318506779161677,
+    "producer_batch_seconds": 0.05445296664061061,
+    "producer_p95_seconds": 0.15581639900028676,
+    "gpu_update_seconds": 0.9665036825890879,
+    "producer_batches_per_second": 18.364472345464602,
+    "gpu_batches_per_second": 1.0346572061900288,
+    "producer_headroom": 17.749330150697002,
     "sustainable": true,
-    "buffer_seconds": 60.75024137519558,
+    "buffer_seconds": 60.88973200311254,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1790696019647648547-3fdea0a8/train.jsonl

@@ -45,3 +45,4 @@
 {"step": 279500, "loss": 0.597193019464612, "lr": 0.0003, "grad_norm": 0.1399536693597878, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40911096080, "tokens_per_second": 149875.69336045466, "data_wait_s": 1.2569000318762846e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 279600, "loss": 0.6400405652821064, "lr": 0.0003, "grad_norm": 0.15598653516577976, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40925686562, "tokens_per_second": 150712.69561718983, "data_wait_s": 1.082600010704482e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 279700, "loss": 0.6038200240582228, "lr": 0.0003, "grad_norm": 0.14256179918893477, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40940438122, "tokens_per_second": 151937.09150502417, "data_wait_s": 7.476999599020928e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 279800, "loss": 0.631115909665823, "lr": 0.0003, "grad_norm": 0.14944488747335424, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40955163321, "tokens_per_second": 150272.22707045428, "data_wait_s": 1.2073000107193366e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790696019.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.22407.0.1790696019647648547-3fdea0a8

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1aecfa60774606494763d42f72090a5f97e116c430cce5fe93fe7918745725c0
-size 99711
+oid sha256:017b0d9c97b8216372b9f311761c5b19a6cb6f47fa181d5eb5f8cab7fe68909a
+size 101829