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": 77472.04557409624,
-      "required_pages_per_second_upper_estimate": 0.005359695872727792
+      "required_token_positions_per_second": 77242.24446080196,
+      "required_pages_per_second_upper_estimate": 0.005343797698498049
     },
     {
       "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": 9684.00569676203,
-      "required_pages_per_second_upper_estimate": 0.004054490960199538
+      "required_token_positions_per_second": 9655.280557600245,
+      "required_pages_per_second_upper_estimate": 0.004042464344281615
     },
     {
       "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": 19368.01139352406,
-      "required_pages_per_second_upper_estimate": 0.0040338460044628995
+      "required_token_positions_per_second": 19310.56111520049,
+      "required_pages_per_second_upper_estimate": 0.004021880626553823
     },
     {
       "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": 29052.017090286092,
-      "required_pages_per_second_upper_estimate": 0.0017416240647010737
+      "required_token_positions_per_second": 28965.841672800732,
+      "required_pages_per_second_upper_estimate": 0.00173645798000507
     },
     {
       "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": 29052.017090286092,
-      "required_pages_per_second_upper_estimate": 0.002258666542451438
+      "required_token_positions_per_second": 28965.841672800732,
+      "required_pages_per_second_upper_estimate": 0.0022519667827875537
     },
     {
       "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": 29052.017090286092,
-      "required_pages_per_second_upper_estimate": 0.3726480816085747
+      "required_token_positions_per_second": 28965.841672800732,
+      "required_pages_per_second_upper_estimate": 0.37154271588102683
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.4230697477346439,
-    "producer_p95_seconds": 0.3381546989985509,
-    "gpu_update_seconds": 0.951672302617654,
-    "producer_batches_per_second": 2.3636764513524517,
-    "gpu_batches_per_second": 1.050781868138241,
-    "producer_headroom": 2.249445411101713,
+    "producer_batch_seconds": 0.09331058877381793,
+    "producer_p95_seconds": 0.3408043550007278,
+    "gpu_update_seconds": 0.9545035946931174,
+    "producer_batches_per_second": 10.716897333313051,
+    "gpu_batches_per_second": 1.0476649910590543,
+    "producer_headroom": 10.22931702860439,
     "sustainable": true,
-    "buffer_seconds": 60.90702736752986,
+    "buffer_seconds": 60.1337264656664,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 279637.8219531567,
-    "upload_bytes_per_second": 5107653.169066347,
+    "generation_bytes_per_second": 278423.45308226545,
+    "upload_bytes_per_second": 5012085.309293936,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.0037669558146,
+    "estimated_full_shards_per_hour": 59.743191665181854,
     "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": 1.2461040010020952,
-      "bytes": 11349081,
-      "pages": 37,
-      "shards": 4
-    },
     {
       "seconds": 1.0920391100007691,
       "bytes": 1249324,
@@ -562,6 +556,12 @@
       "bytes": 10740667,
       "pages": 33,
       "shards": 3
+    },
+    {
+      "seconds": 0.8488108419987839,
+      "bytes": 1104603,
+      "pages": 36,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -359,3 +359,4 @@
 {"step": 593000, "loss": 0.5732935070991516, "lr": 0.0003, "grad_norm": 0.1371412965224625, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86797405638, "tokens_per_second": 141806.3136791834, "data_wait_s": 1.210100163007155e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 593100, "loss": 0.5730262994766235, "lr": 0.0003, "grad_norm": 0.16897583731156532, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86812053714, "tokens_per_second": 152330.1380773663, "data_wait_s": 9.63400088949129e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 593200, "loss": 0.6198545098304749, "lr": 0.0003, "grad_norm": 0.16725111607689697, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86826753976, "tokens_per_second": 153610.33405605968, "data_wait_s": 1.1969001207035035e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 593300, "loss": 0.5967822074890137, "lr": 0.0003, "grad_norm": 0.1605469513533806, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86841502857, "tokens_per_second": 152921.5908893095, "data_wait_s": 1.5127996448427439e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:dad296284536ac5a0a4a7f7ab8d935f9e67a8d440b9b0ff1a09fbe7601369376
-size 770244
+oid sha256:d6358e804bc3d155492bde7f9be7e34b178d08dc905c68061d2911f34a353369
+size 772362