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": 75309.39956736963,
-      "required_pages_per_second_upper_estimate": 0.006527627884571414
+      "required_token_positions_per_second": 75164.30339030511,
+      "required_pages_per_second_upper_estimate": 0.00651505131568636
     },
     {
       "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": 9413.674945921204,
-      "required_pages_per_second_upper_estimate": 0.0037552556829109638
+      "required_token_positions_per_second": 9395.537923788139,
+      "required_pages_per_second_upper_estimate": 0.0037480205536094377
     },
     {
       "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": 18827.34989184241,
-      "required_pages_per_second_upper_estimate": 0.004355376047436654
+      "required_token_positions_per_second": 18791.075847576278,
+      "required_pages_per_second_upper_estimate": 0.00434698468569706
     },
     {
       "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": 28241.02483776361,
-      "required_pages_per_second_upper_estimate": 0.0017197302434135996
+      "required_token_positions_per_second": 28186.61377136442,
+      "required_pages_per_second_upper_estimate": 0.001716416894942692
     },
     {
       "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": 28241.02483776361,
-      "required_pages_per_second_upper_estimate": 0.0021740722681712903
+      "required_token_positions_per_second": 28186.61377136442,
+      "required_pages_per_second_upper_estimate": 0.002169883553660409
     },
     {
       "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": 28241.02483776361,
-      "required_pages_per_second_upper_estimate": 0.3506590118549686
+      "required_token_positions_per_second": 28186.61377136442,
+      "required_pages_per_second_upper_estimate": 0.34998340851241566
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.09496507446087321,
-    "producer_p95_seconds": 0.31303491000289796,
-    "gpu_update_seconds": 0.979001298955319,
-    "producer_batches_per_second": 10.530187078534988,
-    "gpu_batches_per_second": 1.0214491043751306,
-    "producer_headroom": 10.309066828128268,
+    "producer_batch_seconds": 0.33660650432835837,
+    "producer_p95_seconds": 0.3290676649994566,
+    "gpu_update_seconds": 0.9808911501135474,
+    "producer_batches_per_second": 2.970827916695584,
+    "gpu_batches_per_second": 1.0194811115221505,
+    "producer_headroom": 2.9140588119969655,
     "sustainable": true,
-    "buffer_seconds": 60.69808053522978,
+    "buffer_seconds": 60.815251307039944,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 488058.0945348453,
+    "generation_bytes_per_second": 432392.84877908183,
     "upload_bytes_per_second": 2632518.178968874,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 104.7259056762125,
+    "estimated_full_shards_per_hour": 92.78143975762693,
     "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-v1/1790703713325746110-adad6d18/train.jsonl

@@ -220,3 +220,4 @@
 {"step": 302200, "loss": 0.6018653325736523, "lr": 0.0003, "grad_norm": 0.15321024691705712, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44234566710, "tokens_per_second": 151109.95503645364, "data_wait_s": 1.2814001820515841e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 302300, "loss": 0.6214968990534544, "lr": 0.0003, "grad_norm": 0.1450737801302049, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44249181630, "tokens_per_second": 149865.42864868062, "data_wait_s": 1.255000097444281e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 302400, "loss": 0.6266440860927105, "lr": 0.0003, "grad_norm": 0.14723983774197877, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44263812805, "tokens_per_second": 149695.03002851832, "data_wait_s": 8.864000847097486e-06, "gpu_peak_gb": 85.46991729736328}
+{"step": 302500, "loss": 0.5735919512808323, "lr": 0.0003, "grad_norm": 0.14081080755873032, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44278319407, "tokens_per_second": 148182.60438011363, "data_wait_s": 8.684000931680202e-06, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d3c5f81e433a0d6a4a86e0aacf2cbf29bca551230e0939eb34fd9077a869f751
-size 473493
+oid sha256:04e46ea121edff9b60114c9a24c0f8a266f2e844dff2480cdbf314868f89a3c2
+size 475611