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": 76376.29869854441,
-      "required_pages_per_second_upper_estimate": 0.0052838895613940805
+      "required_token_positions_per_second": 74799.7630102293,
+      "required_pages_per_second_upper_estimate": 0.005174821164409661
     },
     {
       "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": 9547.037337318052,
-      "required_pages_per_second_upper_estimate": 0.003997145168324937
+      "required_token_positions_per_second": 9349.970376278663,
+      "required_pages_per_second_upper_estimate": 0.0039146373469638495
     },
     {
       "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": 19094.074674636104,
-      "required_pages_per_second_upper_estimate": 0.003976792210115622
+      "required_token_positions_per_second": 18699.940752557326,
+      "required_pages_per_second_upper_estimate": 0.0038947045081571046
     },
     {
       "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": 28641.112011954156,
-      "required_pages_per_second_upper_estimate": 0.0017169909326708002
+      "required_token_positions_per_second": 28049.911128835993,
+      "required_pages_per_second_upper_estimate": 0.001681549342439346
     },
     {
       "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": 28641.112011954156,
-      "required_pages_per_second_upper_estimate": 0.0022267204799915597
+      "required_token_positions_per_second": 28049.911128835993,
+      "required_pages_per_second_upper_estimate": 0.0021807572117469864
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28641.112011954156,
-      "required_pages_per_second_upper_estimate": 0.3673774324592316
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28049.911128835993,
+      "required_pages_per_second_upper_estimate": 0.3597941423126434
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.0903215247265905,
-    "producer_p95_seconds": 0.26998096800161875,
-    "gpu_update_seconds": 0.9653256475677462,
-    "producer_batches_per_second": 11.07155800377672,
-    "gpu_batches_per_second": 1.0359198499694067,
-    "producer_headroom": 10.687658899579626,
+    "producer_batch_seconds": 0.18188173771090987,
+    "producer_p95_seconds": 0.2642654360024608,
+    "gpu_update_seconds": 0.985671572113367,
+    "producer_batches_per_second": 5.498078106057245,
+    "gpu_batches_per_second": 1.0145367161760703,
+    "producer_headroom": 5.419299290399528,
     "sustainable": true,
-    "buffer_seconds": 60.81551579676801,
+    "buffer_seconds": 60.12596589891539,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 531648.822191744,
+    "generation_bytes_per_second": 523141.29528551653,
     "upload_bytes_per_second": 5169340.789418813,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 114.07946109117736,
+    "estimated_full_shards_per_hour": 112.2539438621914,
     "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-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -219,3 +219,4 @@
 {"step": 741900, "loss": 0.549426257610321, "lr": 5e-05, "grad_norm": 0.23333403917219317, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108593877800, "tokens_per_second": 148868.92985597448, "data_wait_s": 1.5596000594086945e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 742000, "loss": 0.5478297472000122, "lr": 5e-05, "grad_norm": 0.22453109765055274, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108608357372, "tokens_per_second": 149433.93791050083, "data_wait_s": 1.4843000826658681e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 742100, "loss": 0.5540927648544312, "lr": 5e-05, "grad_norm": 0.2346659712796483, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108622933641, "tokens_per_second": 149812.94862111897, "data_wait_s": 1.634900036151521e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 742200, "loss": 0.5303184986114502, "lr": 5e-05, "grad_norm": 0.23299460527597712, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108637634533, "tokens_per_second": 151946.86827446392, "data_wait_s": 1.3856999430572614e-05, "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 sha256:3594e23642755d1f22301fac477dd6e2171952ce87d466062377729cc2571b58
-size 471375
+oid sha256:bd511f296416c785f8d1741ac0423aaf1be45ebfe8d4bd9294a24dc53187ef27
+size 473493