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": 77767.95395949586,
-      "required_pages_per_second_upper_estimate": 0.0053801675014317875
+      "required_token_positions_per_second": 77301.27790435737,
+      "required_pages_per_second_upper_estimate": 0.005347881769099673
     },
     {
       "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": 9720.994244936983,
-      "required_pages_per_second_upper_estimate": 0.004069977292911671
+      "required_token_positions_per_second": 9662.659738044671,
+      "required_pages_per_second_upper_estimate": 0.004045553853038886
     },
     {
       "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": 19441.988489873966,
-      "required_pages_per_second_upper_estimate": 0.0040492534827253615
+      "required_token_positions_per_second": 19325.319476089342,
+      "required_pages_per_second_upper_estimate": 0.004024954403922822
     },
     {
       "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": 29162.982734810947,
-      "required_pages_per_second_upper_estimate": 0.00174827628565561
+      "required_token_positions_per_second": 28987.97921413401,
+      "required_pages_per_second_upper_estimate": 0.0017377850918059317
     },
     {
       "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": 29162.982734810947,
-      "required_pages_per_second_upper_estimate": 0.0022672936332269575
+      "required_token_positions_per_second": 28987.97921413401,
+      "required_pages_per_second_upper_estimate": 0.002253687879253463
     },
     {
       "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": 29162.982734810947,
-      "required_pages_per_second_upper_estimate": 0.37407142975091323
+      "required_token_positions_per_second": 28987.97921413401,
+      "required_pages_per_second_upper_estimate": 0.3718266724918102
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,20 +155,20 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.05780837724219268,
-    "producer_p95_seconds": 0.13653826799986746,
-    "gpu_update_seconds": 0.9480511733457717,
-    "producer_batches_per_second": 17.29853089994937,
-    "gpu_batches_per_second": 1.0547953824801413,
-    "producer_headroom": 16.39989251685509,
+    "producer_batch_seconds": 0.0411862433749981,
+    "producer_p95_seconds": 0.11618356999997559,
+    "gpu_update_seconds": 0.9537746593429092,
+    "producer_batches_per_second": 24.27995170365659,
+    "gpu_batches_per_second": 1.0484656833815833,
+    "producer_headroom": 23.15760266501735,
     "sustainable": true,
-    "buffer_seconds": 60.67527509412939,
+    "buffer_seconds": 60.087803538603275,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 50,
+    "upload_pages_per_commit": 56,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1791033297667653750-ba4975eb/train.jsonl

@@ -4,3 +4,4 @@
 {"step": 600400, "loss": 0.5704043507575989, "lr": 0.0003, "grad_norm": 0.16691128189004883, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87880667765, "tokens_per_second": 152431.60604831862, "data_wait_s": 1.5606000033585588e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 600500, "loss": 0.5799246430397034, "lr": 0.0003, "grad_norm": 0.18639385467417846, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87895218915, "tokens_per_second": 153047.72347827797, "data_wait_s": 1.4684999996461556e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 600600, "loss": 0.5803067684173584, "lr": 0.0003, "grad_norm": 0.17681801482845147, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87909912683, "tokens_per_second": 152818.26869542152, "data_wait_s": 2.0143000028838287e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 600700, "loss": 0.5805760025978088, "lr": 0.0003, "grad_norm": 0.19626544568017945, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87924785186, "tokens_per_second": 156109.97371656948, "data_wait_s": 1.483399978496891e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791033297.83ac8725-d818-41af-bbb1-78dcd88792cf-7zwhs.737.0.1791033297667653750-ba4975eb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d57c9091fedad63e99c2348008486b10b6b879c2901dbce99c02c393011a31a1
-size 12873
+oid sha256:0e10d6f64349c5fc1e2ec7857670f8629a87cdbee3d1f848be26033bf1c9ea92
+size 14991