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": 75834.5279324013,
-      "required_pages_per_second_upper_estimate": 0.006573144680326685
+      "required_token_positions_per_second": 75641.10831783692,
+      "required_pages_per_second_upper_estimate": 0.006556379558353776
     },
     {
       "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": 9479.315991550162,
-      "required_pages_per_second_upper_estimate": 0.003781440877433446
+      "required_token_positions_per_second": 9455.138539729614,
+      "required_pages_per_second_upper_estimate": 0.003771796130417111
     },
     {
       "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": 18958.631983100324,
-      "required_pages_per_second_upper_estimate": 0.004385745848763279
+      "required_token_positions_per_second": 18910.27707945923,
+      "required_pages_per_second_upper_estimate": 0.00437455979282315
     },
     {
       "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": 28437.947974650484,
-      "required_pages_per_second_upper_estimate": 0.0017317218292741592
+      "required_token_positions_per_second": 28365.415619188843,
+      "required_pages_per_second_upper_estimate": 0.001727304989374404
     },
     {
       "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": 28437.947974650484,
-      "required_pages_per_second_upper_estimate": 0.002189231956366974
+      "required_token_positions_per_second": 28365.415619188843,
+      "required_pages_per_second_upper_estimate": 0.0021836482148611266
     },
     {
       "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": 28437.947974650484,
-      "required_pages_per_second_upper_estimate": 0.3531041381557605
+      "required_token_positions_per_second": 28365.415619188843,
+      "required_pages_per_second_upper_estimate": 0.352203529051105
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.07116053917954446,
-    "producer_p95_seconds": 0.25304701999993995,
-    "gpu_update_seconds": 0.9722220472674525,
-    "producer_batches_per_second": 14.052732195815855,
-    "gpu_batches_per_second": 1.028571613666467,
-    "producer_headroom": 13.662376065117336,
+    "producer_batch_seconds": 0.12625771396895402,
+    "producer_p95_seconds": 0.31309941799918306,
+    "gpu_update_seconds": 0.9747080871713538,
+    "producer_batches_per_second": 7.920308142486198,
+    "gpu_batches_per_second": 1.025948192244967,
+    "producer_headroom": 7.719988399370421,
     "sustainable": true,
-    "buffer_seconds": 60.27776693058206,
+    "buffer_seconds": 60.431901404623936,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 290733.70930031873,
+    "generation_bytes_per_second": 289123.97033876705,
     "upload_bytes_per_second": 5174885.176624713,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.384686081477845,
+    "estimated_full_shards_per_hour": 62.039273573134025,
     "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-v1/1790563279934888633-82973089/train.jsonl

@@ -187,3 +187,4 @@
 {"step": 173700, "loss": 0.5771032311022282, "lr": 0.0003, "grad_norm": 0.15708337584257606, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25422283775, "tokens_per_second": 136087.01565575678, "data_wait_s": 1.3551998563343659e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 173800, "loss": 0.5935089848935604, "lr": 0.0003, "grad_norm": 0.14175006275046448, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25436732078, "tokens_per_second": 147430.50651803036, "data_wait_s": 1.1625001206994057e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 173900, "loss": 0.5828830786049366, "lr": 0.0003, "grad_norm": 0.14228330439659898, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25451268564, "tokens_per_second": 148178.62506568458, "data_wait_s": 1.5075002011144534e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 174000, "loss": 0.5894587468355894, "lr": 0.0003, "grad_norm": 0.15220324222693843, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 25465872070, "tokens_per_second": 148222.33531019682, "data_wait_s": 1.2658998457482085e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5ffc70a4481d13cb0c1e1c7ce72b1e840c40ef3a13b3366e96b3d4d34d564a4c
-size 402816
+oid sha256:90f0a1c63df3e40f9ff5a9eae6a6b6fe955b347afc6b23d1354987e7fa333e56
+size 404934