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": 76690.20252291462,
-      "required_pages_per_second_upper_estimate": 0.005305606156321232
+      "required_token_positions_per_second": 76730.7588410305,
+      "required_pages_per_second_upper_estimate": 0.005308411936512112
     },
     {
       "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": 9586.275315364328,
-      "required_pages_per_second_upper_estimate": 0.004013573290350756
+      "required_token_positions_per_second": 9591.344855128813,
+      "required_pages_per_second_upper_estimate": 0.004015695800786118
     },
     {
       "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": 19172.550630728656,
-      "required_pages_per_second_upper_estimate": 0.003993136682219567
+      "required_token_positions_per_second": 19182.689710257626,
+      "required_pages_per_second_upper_estimate": 0.003995248385099943
     },
     {
       "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": 28758.825946092988,
-      "required_pages_per_second_upper_estimate": 0.001724047703283653
+      "required_token_positions_per_second": 28774.034565386435,
+      "required_pages_per_second_upper_estimate": 0.001724959436788078
     },
     {
       "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": 28758.825946092988,
-      "required_pages_per_second_upper_estimate": 0.002235872220601977
+      "required_token_positions_per_second": 28774.034565386435,
+      "required_pages_per_second_upper_estimate": 0.002237054623856396
     },
     {
       "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": 28758.825946092988,
-      "required_pages_per_second_upper_estimate": 0.3688873404149894
+      "required_token_positions_per_second": 28774.034565386435,
+      "required_pages_per_second_upper_estimate": 0.3690824202535426
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09968875856239379,
-    "producer_p95_seconds": 0.27569104700160096,
-    "gpu_update_seconds": 0.961374433428709,
-    "producer_batches_per_second": 10.031221317437854,
-    "gpu_batches_per_second": 1.0401774430733863,
-    "producer_headroom": 9.643759710649805,
+    "producer_batch_seconds": 0.14061521882817374,
+    "producer_p95_seconds": 0.28318861099978676,
+    "gpu_update_seconds": 0.9608662955197463,
+    "producer_batches_per_second": 7.111605758847203,
+    "gpu_batches_per_second": 1.04072752334297,
+    "producer_headroom": 6.833302280700407,
     "sustainable": true,
-    "buffer_seconds": 60.566589306008666,
+    "buffer_seconds": 60.53457661774402,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 312407.3424411557,
-    "upload_bytes_per_second": 4691018.053773159,
+    "generation_bytes_per_second": 313232.8764052069,
+    "upload_bytes_per_second": 4589739.18727165,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.03534321714405,
+    "estimated_full_shards_per_hour": 67.2124835883823,
     "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": [
-    {
-      "seconds": 1.4946176230005221,
-      "bytes": 11950661,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.2175075589984772,
       "bytes": 2434158,
@@ -562,6 +556,12 @@
       "bytes": 10858764,
       "pages": 40,
       "shards": 3
+    },
+    {
+      "seconds": 1.2862951650022296,
+      "bytes": 2384330,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -285,3 +285,4 @@
 {"step": 713500, "loss": 0.5403037667274475, "lr": 8e-05, "grad_norm": 0.21171885586376152, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104436542630, "tokens_per_second": 151992.42126358376, "data_wait_s": 1.5041001461213455e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 713600, "loss": 0.5394705533981323, "lr": 8e-05, "grad_norm": 0.2210485086513882, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104451191709, "tokens_per_second": 152133.34536465406, "data_wait_s": 1.6529997083125636e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 713700, "loss": 0.6320292949676514, "lr": 8e-05, "grad_norm": 0.39268109388291367, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104465893600, "tokens_per_second": 153156.06101313586, "data_wait_s": 1.5458997950190678e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 713800, "loss": 0.5295023322105408, "lr": 8e-05, "grad_norm": 0.21661962613630112, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104480523940, "tokens_per_second": 152592.07676565228, "data_wait_s": 1.0264000593451783e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b87748bc8f76a489df79cadeb59b3d9ea4ed06f15390ab29861a82a1d55714d7
-size 611946
+oid sha256:0c74757487e5ae03dd8095385e2b78272af834a7d78939bcf5ed2b7552226b9e
+size 614064