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": 76804.93691407517,
-      "required_pages_per_second_upper_estimate": 0.005313543747722183
+      "required_token_positions_per_second": 76954.50835702986,
+      "required_pages_per_second_upper_estimate": 0.005323891447199351
     },
     {
       "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": 9600.617114259396,
-      "required_pages_per_second_upper_estimate": 0.004019577902057304
+      "required_token_positions_per_second": 9619.313544628732,
+      "required_pages_per_second_upper_estimate": 0.004027405707027082
     },
     {
       "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": 19201.23422851879,
-      "required_pages_per_second_upper_estimate": 0.00399911071920191
+      "required_token_positions_per_second": 19238.627089257465,
+      "required_pages_per_second_upper_estimate": 0.004006898665977725
     },
     {
       "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": 28801.851342778184,
-      "required_pages_per_second_upper_estimate": 0.0017266270101189535
+      "required_token_positions_per_second": 28857.9406338862,
+      "required_pages_per_second_upper_estimate": 0.0017299894774774904
     },
     {
       "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": 28801.851342778184,
-      "required_pages_per_second_upper_estimate": 0.002239217255945531
+      "required_token_positions_per_second": 28857.9406338862,
+      "required_pages_per_second_upper_estimate": 0.002243577951618461
     },
     {
       "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": 28801.851342778184,
-      "required_pages_per_second_upper_estimate": 0.3694392240065954
+      "required_token_positions_per_second": 28857.9406338862,
+      "required_pages_per_second_upper_estimate": 0.37015867720894036
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.09882597342186727,
-    "producer_p95_seconds": 0.2954445060022408,
-    "gpu_update_seconds": 0.9599382925407848,
-    "producer_batches_per_second": 10.118797370517267,
-    "gpu_batches_per_second": 1.0417336278493268,
-    "producer_headroom": 9.713421070420528,
+    "producer_batch_seconds": 0.07867285049997008,
+    "producer_p95_seconds": 0.275796330999583,
+    "gpu_update_seconds": 0.9580725232879079,
+    "producer_batches_per_second": 12.710865230444146,
+    "gpu_batches_per_second": 1.0437623203807218,
+    "producer_headroom": 12.177930724504158,
     "sustainable": true,
-    "buffer_seconds": 60.47611243006944,
+    "buffer_seconds": 60.3585689671382,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 452789.57352674654,
+    "generation_bytes_per_second": 454040.5014910162,
     "upload_bytes_per_second": 6117747.92120753,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 97.15810207702444,
+    "estimated_full_shards_per_hour": 97.42652209804406,
     "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-v1/1790914434605211575-090606d4/train.jsonl

@@ -294,3 +294,4 @@
 {"step": 514400, "loss": 0.5038048624992371, "lr": 0.0003, "grad_norm": 0.26869483123555776, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75290707061, "tokens_per_second": 150772.69595658488, "data_wait_s": 1.7244998161913827e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 514500, "loss": 0.5570145845413208, "lr": 0.0003, "grad_norm": 0.1675595931676757, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75305303347, "tokens_per_second": 152018.73569825958, "data_wait_s": 1.6735997633077204e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 514600, "loss": 0.6139761209487915, "lr": 0.0003, "grad_norm": 0.1643211878055086, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75320020613, "tokens_per_second": 153367.62589523522, "data_wait_s": 1.5787001757416874e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 514700, "loss": 0.6013380885124207, "lr": 0.0003, "grad_norm": 0.15445432625167307, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75334633324, "tokens_per_second": 151803.7379134772, "data_wait_s": 1.4360000932356343e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b78a50cb1876c4b07f9d8a6b23939be30eb5b30070882230ec9a0040efd6fa68
-size 631008
+oid sha256:dbcc6ace69610880cfd55358639aa24fb06db3118d87eab844e95397938e11ea
+size 633126