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": 76043.67484817725,
-      "required_pages_per_second_upper_estimate": 0.005260877871632015
+      "required_token_positions_per_second": 75994.36339704735,
+      "required_pages_per_second_upper_estimate": 0.005257466391024517
     },
     {
       "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": 9505.459356022156,
-      "required_pages_per_second_upper_estimate": 0.003979737335803327
+      "required_token_positions_per_second": 9499.295424630918,
+      "required_pages_per_second_upper_estimate": 0.003977156626447339
     },
     {
       "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": 19010.918712044313,
-      "required_pages_per_second_upper_estimate": 0.0039594730160779565
+      "required_token_positions_per_second": 18998.590849261836,
+      "required_pages_per_second_upper_estimate": 0.00395690544736797
     },
     {
       "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": 28516.378068066468,
-      "required_pages_per_second_upper_estimate": 0.0017095133231924376
+      "required_token_positions_per_second": 28497.886273892756,
+      "required_pages_per_second_upper_estimate": 0.0017084047683670585
     },
     {
       "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": 28516.378068066468,
-      "required_pages_per_second_upper_estimate": 0.0022170229645009283
+      "required_token_positions_per_second": 28497.886273892756,
+      "required_pages_per_second_upper_estimate": 0.002215585308840729
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28516.378068066468,
-      "required_pages_per_second_upper_estimate": 0.3657774793559147
+      "required_token_positions_per_second": 28497.886273892756,
+      "required_pages_per_second_upper_estimate": 0.36554028647519604
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.14586033255449138,
-    "producer_p95_seconds": 0.33515112299937755,
-    "gpu_update_seconds": 0.9695480938710479,
-    "producer_batches_per_second": 6.855873577735153,
-    "gpu_batches_per_second": 1.0314083502628208,
-    "producer_headroom": 6.647099159113999,
+    "producer_batch_seconds": 0.09752626219574267,
+    "producer_p95_seconds": 0.3393667609998374,
+    "gpu_update_seconds": 0.9701772171548264,
+    "producer_batches_per_second": 10.253648376197617,
+    "gpu_batches_per_second": 1.0307395209017924,
+    "producer_headroom": 9.94785604730351,
     "sustainable": true,
-    "buffer_seconds": 60.11198182000497,
+    "buffer_seconds": 60.150987463599236,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "shard_target_mib": 32,
+    "upload_pages_per_commit": 170,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 454302.00789400976,
+    "generation_bytes_per_second": 567638.4694907954,
     "upload_bytes_per_second": 4489788.073212042,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 97.48263528456897,
+    "estimated_full_shards_per_hour": 60.901000802721484,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -428,3 +428,4 @@
 {"step": 802800, "loss": 0.545868456363678, "lr": 3e-05, "grad_norm": 0.2683358329979781, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117506774323, "tokens_per_second": 152528.6755177442, "data_wait_s": 1.4285993529483676e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 802900, "loss": 0.5391790270805359, "lr": 3e-05, "grad_norm": 0.2842515022638538, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117521292308, "tokens_per_second": 150410.5781571794, "data_wait_s": 2.0007006241939962e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 803000, "loss": 0.5340162515640259, "lr": 3e-05, "grad_norm": 0.274242308446818, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117535845279, "tokens_per_second": 151410.89411223173, "data_wait_s": 1.251199864782393e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 803100, "loss": 0.5561849474906921, "lr": 3e-05, "grad_norm": 0.2756510483160743, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117550351913, "tokens_per_second": 151219.05238726872, "data_wait_s": 1.7641999875195324e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:95273356fc9ce8d904868f7140dfe4fa5294abb5f3bd9e661aee333bf8477219
-size 917169
+oid sha256:fb10462a1506a1e7ae9fd457f5ee3ba8aab876a53c82b6094200531a58d20473
+size 919287