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": 74995.73895901168,
-      "required_pages_per_second_upper_estimate": 0.00650044057800449
+      "required_token_positions_per_second": 75531.63849836394,
+      "required_pages_per_second_upper_estimate": 0.00654689099185058
     },
     {
       "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": 9374.46736987646,
-      "required_pages_per_second_upper_estimate": 0.0037396151946212145
+      "required_token_positions_per_second": 9441.454812295493,
+      "required_pages_per_second_upper_estimate": 0.0037663374869536833
     },
     {
       "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": 18748.93473975292,
-      "required_pages_per_second_upper_estimate": 0.004337236082060293
+      "required_token_positions_per_second": 18882.909624590986,
+      "required_pages_per_second_upper_estimate": 0.004368228813790138
     },
     {
       "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": 28123.40210962938,
-      "required_pages_per_second_upper_estimate": 0.0017125676363889861
+      "required_token_positions_per_second": 28324.364436886477,
+      "required_pages_per_second_upper_estimate": 0.0017248051877510981
     },
     {
       "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": 28123.40210962938,
-      "required_pages_per_second_upper_estimate": 0.0021650173449589668
+      "required_token_positions_per_second": 28324.364436886477,
+      "required_pages_per_second_upper_estimate": 0.0021804879812105447
     },
     {
       "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": 28123.40210962938,
-      "required_pages_per_second_upper_estimate": 0.3491985312294893
+      "required_token_positions_per_second": 28324.364436886477,
+      "required_pages_per_second_upper_estimate": 0.3516938107563788
     }
   ],
   "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": 0.04127467762488379,
-    "producer_p95_seconds": 0.1285456149998936,
-    "gpu_update_seconds": 0.9830958534896955,
-    "producer_batches_per_second": 24.227929993500837,
-    "gpu_batches_per_second": 1.0171948100994423,
-    "producer_headroom": 23.8183775152493,
+    "producer_batch_seconds": 0.039286027031295134,
+    "producer_p95_seconds": 0.12814575899938063,
+    "gpu_update_seconds": 0.9761207550342893,
+    "producer_batches_per_second": 25.454342817699608,
+    "gpu_batches_per_second": 1.0244634127924797,
+    "producer_headroom": 24.84651233011458,
     "sustainable": true,
-    "buffer_seconds": 60.95194291636112,
+    "buffer_seconds": 60.51948681212593,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

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

@@ -22,3 +22,4 @@
 {"step": 87200, "loss": 0.583887867629528, "lr": 0.0003, "grad_norm": 0.13447308791749443, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12761166726, "tokens_per_second": 150569.09252748732, "data_wait_s": 8.68499955686275e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 87300, "loss": 0.6145248562097549, "lr": 0.0003, "grad_norm": 0.13674184973775003, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12775735680, "tokens_per_second": 148645.33195751367, "data_wait_s": 7.156000719987787e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 87400, "loss": 0.6537573337554932, "lr": 0.0003, "grad_norm": 0.13270455503542447, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12790268643, "tokens_per_second": 148972.17118350157, "data_wait_s": 1.534100010758266e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 87500, "loss": 0.6125136204063892, "lr": 0.0003, "grad_norm": 0.13232944352760084, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12804833533, "tokens_per_second": 149206.4226659045, "data_wait_s": 8.506000085617416e-06, "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 sha256:085016114e1e013c9d5d087cd6d6446257a463204749e7e308f24a3ac9ff05f7
-size 50997
+oid sha256:e14823e6c59a3bc962906625e4c8d4d57cf8af9df4dd80cd5cf456673f56a3cc
+size 53115