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": 77528.85116441638,
-      "required_pages_per_second_upper_estimate": 0.0053636258152732685
+      "required_token_positions_per_second": 77419.45324774673,
+      "required_pages_per_second_upper_estimate": 0.005356057413559923
     },
     {
       "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": 9691.106395552048,
-      "required_pages_per_second_upper_estimate": 0.004057463874503467
+      "required_token_positions_per_second": 9677.431655968341,
+      "required_pages_per_second_upper_estimate": 0.004051738546600804
     },
     {
       "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": 19382.212791104095,
-      "required_pages_per_second_upper_estimate": 0.0040368037810627816
+      "required_token_positions_per_second": 19354.863311936682,
+      "required_pages_per_second_upper_estimate": 0.004031107605806478
     },
     {
       "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": 29073.319186656143,
-      "required_pages_per_second_upper_estimate": 0.0017429010928520403
+      "required_token_positions_per_second": 29032.294967905025,
+      "required_pages_per_second_upper_estimate": 0.0017404417535782659
     },
     {
       "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": 29073.319186656143,
-      "required_pages_per_second_upper_estimate": 0.00226032268674619
+      "required_token_positions_per_second": 29032.294967905025,
+      "required_pages_per_second_upper_estimate": 0.0022571332341847585
     },
     {
       "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": 29073.319186656143,
-      "required_pages_per_second_upper_estimate": 0.37292132202840067
+      "required_token_positions_per_second": 29032.294967905025,
+      "required_pages_per_second_upper_estimate": 0.3723951073986355
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,17 +158,17 @@
     "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.047619281101569655,
-    "producer_p95_seconds": 0.12511030299992854,
-    "gpu_update_seconds": 0.9509750098533529,
-    "producer_batches_per_second": 20.999897034712635,
-    "gpu_batches_per_second": 1.0515523432673664,
-    "producer_headroom": 19.970377289505244,
+    "producer_batch_seconds": 0.05295030008592683,
+    "producer_p95_seconds": 0.12384743999996317,
+    "gpu_update_seconds": 0.9523187894916557,
+    "producer_batches_per_second": 18.88563423393668,
+    "gpu_batches_per_second": 1.0500685390590647,
+    "producer_headroom": 17.98514433244475,
     "sustainable": true,
-    "buffer_seconds": 60.86240063061459,
+    "buffer_seconds": 60.948402527465966,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 62,
+    "upload_pages_per_commit": 63,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -6,3 +6,4 @@
 {"step": 685600, "loss": 0.5076792240142822, "lr": 8e-05, "grad_norm": 0.19537428987943656, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100354050683, "tokens_per_second": 151189.80668697378, "data_wait_s": 2.0197000139887678e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 685700, "loss": 0.5325490832328796, "lr": 8e-05, "grad_norm": 0.20408287366589967, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100368548106, "tokens_per_second": 152280.91207102733, "data_wait_s": 1.5908000023046043e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 685800, "loss": 0.5718104243278503, "lr": 8e-05, "grad_norm": 0.21553502583604203, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100383214867, "tokens_per_second": 153879.53710993732, "data_wait_s": 1.6706999986126903e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 685900, "loss": 0.5103606581687927, "lr": 8e-05, "grad_norm": 0.20037664165679023, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100397882630, "tokens_per_second": 153820.61905108768, "data_wait_s": 1.2389000175971887e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:105a4b35eec04747f1616147bbd70040906c10a8206a217f602fd04be0b52292
-size 17109
+oid sha256:04ea5be46f0cb89eb70ada72a40066016d25c8dca711dfed7427c45339df48e1
+size 19227