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": 76931.96642289756,
-      "required_pages_per_second_upper_estimate": 0.005322331943891586
+      "required_token_positions_per_second": 77306.44531639626,
+      "required_pages_per_second_upper_estimate": 0.005348239262654571
     },
     {
       "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": 9616.495802862195,
-      "required_pages_per_second_upper_estimate": 0.004026225977390572
+      "required_token_positions_per_second": 9663.305664549533,
+      "required_pages_per_second_upper_estimate": 0.0040458242889779925
     },
     {
       "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": 19232.99160572439,
-      "required_pages_per_second_upper_estimate": 0.004005724943375481
+      "required_token_positions_per_second": 19326.611329099065,
+      "required_pages_per_second_upper_estimate": 0.004025223462836292
     },
     {
       "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": 28849.487408586585,
-      "required_pages_per_second_upper_estimate": 0.0017294827195281062
+      "required_token_positions_per_second": 28989.916993648596,
+      "required_pages_per_second_upper_estimate": 0.001737901258729013
     },
     {
       "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": 28849.487408586585,
-      "required_pages_per_second_upper_estimate": 0.002242920750533225
+      "required_token_positions_per_second": 28989.916993648596,
+      "required_pages_per_second_upper_estimate": 0.0022538385330873306
     },
     {
       "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": 28849.487408586585,
-      "required_pages_per_second_upper_estimate": 0.37005024831116307
+      "required_token_positions_per_second": 28989.916993648596,
+      "required_pages_per_second_upper_estimate": 0.3718515282467977
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.13169073731270942,
-    "producer_p95_seconds": 0.2990591170018888,
-    "gpu_update_seconds": 0.9583532493465038,
-    "producer_batches_per_second": 7.593548494040442,
-    "gpu_batches_per_second": 1.0434565758314014,
-    "producer_headroom": 7.277301873333908,
+    "producer_batch_seconds": 0.07946883871886712,
+    "producer_p95_seconds": 0.2974725280000712,
+    "gpu_update_seconds": 0.9537109059697343,
+    "producer_batches_per_second": 12.583548672928886,
+    "gpu_batches_per_second": 1.0485357708929615,
+    "producer_headroom": 12.001067605173256,
     "sustainable": true,
-    "buffer_seconds": 60.37625470882974,
+    "buffer_seconds": 60.08378707609326,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 339937.3388862202,
-    "upload_bytes_per_second": 6132928.734952995,
+    "generation_bytes_per_second": 339463.8225605243,
+    "upload_bytes_per_second": 6096237.535567352,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.94263958873705,
+    "estimated_full_shards_per_hour": 72.84103400813862,
     "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": 1.516792244001408,
-      "bytes": 10349687,
-      "pages": 29,
-      "shards": 3
-    },
     {
       "seconds": 1.0428093720001925,
       "bytes": 2636885,
@@ -562,6 +556,12 @@
       "bytes": 6912505,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.2596074149987544,
+      "bytes": 6204319,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -251,3 +251,4 @@
 {"step": 510100, "loss": 0.5549055933952332, "lr": 0.0003, "grad_norm": 0.14961335480457097, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74661268188, "tokens_per_second": 136093.71024497124, "data_wait_s": 1.6253998182946816e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 510200, "loss": 0.5445055961608887, "lr": 0.0003, "grad_norm": 0.15837737654429215, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74675876472, "tokens_per_second": 152052.4016960982, "data_wait_s": 1.5710997104179114e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 510300, "loss": 0.5971457362174988, "lr": 0.0003, "grad_norm": 0.16460713640988167, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74690558967, "tokens_per_second": 152727.17945273404, "data_wait_s": 1.7277998267672956e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 510400, "loss": 0.5947121381759644, "lr": 0.0003, "grad_norm": 0.14776955372029305, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74705240141, "tokens_per_second": 150967.06811949794, "data_wait_s": 1.730199801386334e-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:e16166f3f3a4dfcc96ea4a501781f83f8acee6541603de2b331f37aa41b9a3e0
-size 539934
+oid sha256:c8f1d45ae435e0cd59f5931af60553d1ca47378d5cca3f11cb94a88e4afaea5f
+size 542052