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": 76803.2631168488,
-      "required_pages_per_second_upper_estimate": 0.005313427950546319
+      "required_token_positions_per_second": 76609.99381876891,
+      "required_pages_per_second_upper_estimate": 0.005300057132058597
     },
     {
       "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": 9600.4078896061,
-      "required_pages_per_second_upper_estimate": 0.0040194903040640766
+      "required_token_positions_per_second": 9576.249227346114,
+      "required_pages_per_second_upper_estimate": 0.004009375576666056
     },
     {
       "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": 19200.8157792122,
-      "required_pages_per_second_upper_estimate": 0.003999023567246598
+      "required_token_positions_per_second": 19152.498454692228,
+      "required_pages_per_second_upper_estimate": 0.003988960342762622
     },
     {
       "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": 28801.2236688183,
-      "required_pages_per_second_upper_estimate": 0.0017265893820234612
+      "required_token_positions_per_second": 28728.747682038338,
+      "required_pages_per_second_upper_estimate": 0.0017222445572804266
     },
     {
       "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": 28801.2236688183,
-      "required_pages_per_second_upper_estimate": 0.0022391684570559965
+      "required_token_positions_per_second": 28728.747682038338,
+      "required_pages_per_second_upper_estimate": 0.0022335337678720296
     },
     {
       "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": 28801.2236688183,
-      "required_pages_per_second_upper_estimate": 0.3694311728789818
+      "required_token_positions_per_second": 28728.747682038338,
+      "required_pages_per_second_upper_estimate": 0.3685015287392201
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.32805258050002806,
+    "producer_batch_seconds": 0.10989997596875867,
     "producer_p95_seconds": 0.19739384300009988,
-    "gpu_update_seconds": 0.9599592127723782,
-    "producer_batches_per_second": 3.0482918271082293,
-    "gpu_batches_per_second": 1.0417109255214951,
-    "producer_headroom": 2.92623582265129,
+    "gpu_update_seconds": 0.9623809678723295,
+    "producer_batches_per_second": 9.09918306337274,
+    "gpu_batches_per_second": 1.0390895428977986,
+    "producer_headroom": 8.756880603376166,
     "sustainable": true,
-    "buffer_seconds": 60.47743040465983,
+    "buffer_seconds": 60.63000097595676,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 354299.2241121041,
-    "upload_bytes_per_second": 4829711.445438087,
+    "generation_bytes_per_second": 367071.5714670663,
+    "upload_bytes_per_second": 4844863.653281563,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.02436583063452,
+    "estimated_full_shards_per_hour": 78.76501424798003,
     "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": [
@@ -496,6 +496,12 @@
       "bytes": 11644292,
       "pages": 29,
       "shards": 3
+    },
+    {
+      "seconds": 1.3249330049984565,
+      "bytes": 7496918,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -103,3 +103,4 @@
 {"step": 730300, "loss": 0.5685901641845703, "lr": 5e-05, "grad_norm": 0.2754684074547258, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106895188806, "tokens_per_second": 150372.91048703482, "data_wait_s": 8.905999493435957e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 730400, "loss": 0.553598165512085, "lr": 5e-05, "grad_norm": 0.22882784487219351, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106909731207, "tokens_per_second": 150398.9126238805, "data_wait_s": 1.2379998224787414e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 730500, "loss": 0.5534103512763977, "lr": 5e-05, "grad_norm": 0.23025847512882885, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106924316428, "tokens_per_second": 150874.42396131923, "data_wait_s": 9.874000170384534e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 730600, "loss": 0.5380862951278687, "lr": 5e-05, "grad_norm": 0.22928137124406728, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106938987568, "tokens_per_second": 150996.0730518774, "data_wait_s": 7.971999366418459e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5586925812a4e35002f331c74ac83c5a22e632e7a31591dc1a02829f0923c238
-size 224121
+oid sha256:b331d9632decf2ae76969be3544577aa82f01a2f3a76d45b183edb1faeb504dc
+size 226239