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": 76845.88990665367,
-      "required_pages_per_second_upper_estimate": 0.005316376970772799
+      "required_token_positions_per_second": 76624.52592155847,
+      "required_pages_per_second_upper_estimate": 0.005301062496648703
     },
     {
       "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": 9605.736238331709,
-      "required_pages_per_second_upper_estimate": 0.004021721172406914
+      "required_token_positions_per_second": 9578.065740194808,
+      "required_pages_per_second_upper_estimate": 0.004010136112662702
     },
     {
       "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": 19211.472476663417,
-      "required_pages_per_second_upper_estimate": 0.004001243076289676
+      "required_token_positions_per_second": 19156.131480389617,
+      "required_pages_per_second_upper_estimate": 0.003989717006206058
     },
     {
       "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": 28817.208714995126,
-      "required_pages_per_second_upper_estimate": 0.0017275476611340096
+      "required_token_positions_per_second": 28734.197220584425,
+      "required_pages_per_second_upper_estimate": 0.001722571248795301
     },
     {
       "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": 28817.208714995126,
-      "required_pages_per_second_upper_estimate": 0.002240411224085457
+      "required_token_positions_per_second": 28734.197220584425,
+      "required_pages_per_second_upper_estimate": 0.002233957445524001
     },
     {
       "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": 23,
-      "required_token_positions_per_second": 28817.208714995126,
-      "required_pages_per_second_upper_estimate": 0.3696362118879328
+      "required_token_positions_per_second": 28734.197220584425,
+      "required_pages_per_second_upper_estimate": 0.3685714295684307
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.05144120615624903,
-    "producer_p95_seconds": 0.11940716799995243,
-    "gpu_update_seconds": 0.9594267187166283,
-    "producer_batches_per_second": 19.439668598799386,
-    "gpu_batches_per_second": 1.0422890883606455,
-    "producer_headroom": 18.650937456684773,
+    "producer_batch_seconds": 0.05057800217969177,
+    "producer_p95_seconds": 0.11771395600010237,
+    "gpu_update_seconds": 0.9621984490380576,
+    "producer_batches_per_second": 19.77144127692578,
+    "gpu_batches_per_second": 1.0392866471565547,
+    "producer_headroom": 19.02405013190502,
     "sustainable": true,
-    "buffer_seconds": 60.443883279147585,
+    "buffer_seconds": 60.61850228939763,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 64,
+    "upload_pages_per_commit": 59,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -4,3 +4,4 @@
 {"step": 720400, "loss": 0.5264872312545776, "lr": 5e-05, "grad_norm": 0.20526975189012117, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105446489981, "tokens_per_second": 153182.13289004244, "data_wait_s": 1.7705999994177546e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 720500, "loss": 0.5409865379333496, "lr": 5e-05, "grad_norm": 0.22716974496486186, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105461179941, "tokens_per_second": 152853.1822943083, "data_wait_s": 1.070100006472785e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 720600, "loss": 0.5312745571136475, "lr": 5e-05, "grad_norm": 0.21549012287903854, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105475828610, "tokens_per_second": 152396.13794055022, "data_wait_s": 1.341500001217355e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 720700, "loss": 0.519378662109375, "lr": 5e-05, "grad_norm": 0.22352734966768664, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105490324667, "tokens_per_second": 150780.89103931733, "data_wait_s": 1.3140999953975552e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:43dbbc84ff2e486ef743998bdac0cebe6528ac51d11cd1f38fe533818a331972
-size 12873
+oid sha256:b4b1de30b2b020dbd6cdc2c366501461e8d94a7aace78dee98a3284abcc08889
+size 14991