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": 76097.54862461527,
-      "required_pages_per_second_upper_estimate": 0.006595942647307943
+      "required_token_positions_per_second": 75492.04733911908,
+      "required_pages_per_second_upper_estimate": 0.00654345933051011
     },
     {
       "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": 9512.193578076909,
-      "required_pages_per_second_upper_estimate": 0.0037945562382626892
+      "required_token_positions_per_second": 9436.505917389884,
+      "required_pages_per_second_upper_estimate": 0.0037643632987832635
     },
     {
       "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": 19024.387156153818,
-      "required_pages_per_second_upper_estimate": 0.004400957150797685
+      "required_token_positions_per_second": 18873.01183477977,
+      "required_pages_per_second_upper_estimate": 0.004365939134312464
     },
     {
       "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": 28536.580734230727,
-      "required_pages_per_second_upper_estimate": 0.0017377280468464998
+      "required_token_positions_per_second": 28309.517752169653,
+      "required_pages_per_second_upper_estimate": 0.0017239011025463782
     },
     {
       "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": 28536.580734230727,
-      "required_pages_per_second_upper_estimate": 0.0021968249792324148
+      "required_token_positions_per_second": 28309.517752169653,
+      "required_pages_per_second_upper_estimate": 0.00217934504232279
     },
     {
       "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": 28536.580734230727,
-      "required_pages_per_second_upper_estimate": 0.3543288269271357
+      "required_token_positions_per_second": 28309.517752169653,
+      "required_pages_per_second_upper_estimate": 0.3515094646208532
     }
   ],
   "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": 0.06525288655468131,
-    "producer_p95_seconds": 0.18386026299958758,
-    "gpu_update_seconds": 0.9688616957123795,
-    "producer_batches_per_second": 15.32499254514985,
-    "gpu_batches_per_second": 1.0321390601211924,
-    "producer_headroom": 14.847798264073457,
+    "producer_batch_seconds": 0.05581754777345793,
+    "producer_p95_seconds": 0.14918788700015284,
+    "gpu_update_seconds": 0.9766326732245746,
+    "producer_batches_per_second": 17.91551295049036,
+    "gpu_batches_per_second": 1.0239264233278955,
+    "producer_headroom": 17.496875305026887,
     "sustainable": true,
-    "buffer_seconds": 60.069425134167524,
+    "buffer_seconds": 60.55122573992362,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 75,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -43,3 +43,4 @@
 {"step": 119300, "loss": 0.6420309375971556, "lr": 0.0003, "grad_norm": 0.13267096332127562, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17457970719, "tokens_per_second": 151675.77809098744, "data_wait_s": 1.3529000170819927e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 119400, "loss": 0.6104239374399185, "lr": 0.0003, "grad_norm": 0.1652067339159888, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17472747252, "tokens_per_second": 151599.0164032128, "data_wait_s": 1.5113000699784607e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 119500, "loss": 0.5577743649482727, "lr": 0.0003, "grad_norm": 0.13060694371417733, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17487473060, "tokens_per_second": 151600.22361494292, "data_wait_s": 1.4679000742034987e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 119600, "loss": 0.6463970169425011, "lr": 0.0003, "grad_norm": 0.15078117921742035, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17502101396, "tokens_per_second": 150354.02502220837, "data_wait_s": 9.141999726125505e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8865b0b615b7dbc8538d584665b961e87d2874d4155c7e25385462f51d18bc05
-size 95475
+oid sha256:e85943d29fb20ef162d9ace3859481938da2486fa211d717979ef275e0aaf9ae
+size 97593