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": 75815.2389856887,
-      "required_pages_per_second_upper_estimate": 0.006571472763313026
+      "required_token_positions_per_second": 74964.46013196302,
+      "required_pages_per_second_upper_estimate": 0.006497729408551375
     },
     {
       "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": 9476.904873211088,
-      "required_pages_per_second_upper_estimate": 0.0037804790462785576
+      "required_token_positions_per_second": 9370.557516495377,
+      "required_pages_per_second_upper_estimate": 0.003738055495649983
     },
     {
       "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": 18953.809746422176,
-      "required_pages_per_second_upper_estimate": 0.004384630309176256
+      "required_token_positions_per_second": 18741.115032990754,
+      "required_pages_per_second_upper_estimate": 0.004335427130523004
     },
     {
       "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": 28430.71461963326,
-      "required_pages_per_second_upper_estimate": 0.001731281355904091
+      "required_token_positions_per_second": 28111.67254948613,
+      "required_pages_per_second_upper_estimate": 0.0017118533677165082
     },
     {
       "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": 28430.71461963326,
-      "required_pages_per_second_upper_estimate": 0.00218867511267454
+      "required_token_positions_per_second": 28111.67254948613,
+      "required_pages_per_second_upper_estimate": 0.0021641143709496386
     },
     {
       "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": 28430.71461963326,
-      "required_pages_per_second_upper_estimate": 0.35301432409492856
+      "required_token_positions_per_second": 28111.67254948613,
+      "required_pages_per_second_upper_estimate": 0.34905288934882267
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,17 +158,17 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.04182291763281398,
-    "producer_p95_seconds": 0.14295848899928387,
-    "gpu_update_seconds": 0.9724694004317167,
-    "producer_batches_per_second": 23.910335686752916,
-    "gpu_batches_per_second": 1.0283099905828001,
-    "producer_headroom": 23.252069809417687,
+    "producer_batch_seconds": 0.03946294715626664,
+    "producer_p95_seconds": 0.13601503300014883,
+    "gpu_update_seconds": 0.9835060490026017,
+    "producer_batches_per_second": 25.34022601100136,
+    "gpu_batches_per_second": 1.016770563855835,
+    "producer_headroom": 24.92226556491291,
     "sustainable": true,
-    "buffer_seconds": 60.293102826766436,
+    "buffer_seconds": 60.977375038161306,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -36,3 +36,4 @@
 {"step": 88600, "loss": 0.6344563355669379, "lr": 0.0003, "grad_norm": 0.13862075684825392, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12966471919, "tokens_per_second": 148368.41707397727, "data_wait_s": 9.277999197365716e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 88700, "loss": 0.6234282646328211, "lr": 0.0003, "grad_norm": 0.13362376395729664, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12980961945, "tokens_per_second": 148160.8179684559, "data_wait_s": 1.4209999790182337e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 88800, "loss": 0.601841427385807, "lr": 0.0003, "grad_norm": 0.14304887042007033, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12995358103, "tokens_per_second": 147155.52727486493, "data_wait_s": 1.3916998796048574e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 88900, "loss": 0.6370924413204193, "lr": 0.0003, "grad_norm": 0.13101865207589453, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13010034356, "tokens_per_second": 149930.09538276383, "data_wait_s": 1.265399987460114e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:38b59442c0cbdad5d0bc18c6f7310ff51803ecc5f4c596d5e7a4438ddcf5e7ce
-size 80649
+oid sha256:79bff12a80104e76bf8dda77af880b3a921d1ecb11f0b20202f7ad3ed2e9d235
+size 82767