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": 76774.29183847796,
-      "required_pages_per_second_upper_estimate": 0.005311423650285997
+      "required_token_positions_per_second": 77336.5422634165,
+      "required_pages_per_second_upper_estimate": 0.00535032143928397
     },
     {
       "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": 9596.786479809745,
-      "required_pages_per_second_upper_estimate": 0.004017974095405979
+      "required_token_positions_per_second": 9667.067782927063,
+      "required_pages_per_second_upper_estimate": 0.0040473994093807
     },
     {
       "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": 19193.57295961949,
-      "required_pages_per_second_upper_estimate": 0.0039975150789314335
+      "required_token_positions_per_second": 19334.135565854125,
+      "required_pages_per_second_upper_estimate": 0.004026790562924904
     },
     {
       "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": 28790.359439429238,
-      "required_pages_per_second_upper_estimate": 0.0017259380880603047
+      "required_token_positions_per_second": 29001.20334878119,
+      "required_pages_per_second_upper_estimate": 0.0017385778584859456
     },
     {
       "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": 28790.359439429238,
-      "required_pages_per_second_upper_estimate": 0.0022383238110076915
+      "required_token_positions_per_second": 29001.20334878119,
+      "required_pages_per_second_upper_estimate": 0.0022547159975554597
     },
     {
       "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": 28790.359439429238,
-      "required_pages_per_second_upper_estimate": 0.3692918182094796
+      "required_token_positions_per_second": 29001.20334878119,
+      "required_pages_per_second_upper_estimate": 0.3719962974920946
     }
   ],
   "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": 0.0800216445000217,
-    "producer_p95_seconds": 0.21884750300159794,
-    "gpu_update_seconds": 0.9603214596249624,
-    "producer_batches_per_second": 12.496618961632674,
-    "gpu_batches_per_second": 1.0413179774099117,
-    "producer_headroom": 12.00077136161207,
+    "producer_batch_seconds": 0.07923231338287451,
+    "producer_p95_seconds": 0.21579302000100142,
+    "gpu_update_seconds": 0.953339751716266,
+    "producer_batches_per_second": 12.621113246658563,
+    "gpu_batches_per_second": 1.0489439868627455,
+    "producer_headroom": 12.03220896895235,
     "sustainable": true,
-    "buffer_seconds": 60.50025195637263,
+    "buffer_seconds": 60.06040435812476,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 487014.7095485847,
-    "upload_bytes_per_second": 5836261.556493634,
+    "generation_bytes_per_second": 470332.9722632305,
+    "upload_bytes_per_second": 5693128.467768653,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 104.50201954692035,
+    "estimated_full_shards_per_hour": 100.92250705645262,
     "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": 1.3153861170003438,
-      "bytes": 17177113,
-      "pages": 40,
-      "shards": 4
-    },
     {
       "seconds": 0.9797320220004622,
       "bytes": 1103154,
@@ -562,6 +556,12 @@
       "bytes": 7163586,
       "pages": 36,
       "shards": 4
+    },
+    {
+      "seconds": 1.230830958000297,
+      "bytes": 6020552,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -135,3 +135,4 @@
 {"step": 538500, "loss": 0.5491064786911011, "lr": 0.0003, "grad_norm": 0.15972445504468055, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78818488987, "tokens_per_second": 153061.73812530036, "data_wait_s": 1.0843999916687608e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 538600, "loss": 0.5387080311775208, "lr": 0.0003, "grad_norm": 0.16702672484838227, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78833117219, "tokens_per_second": 152753.8111222553, "data_wait_s": 1.3556998965214007e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 538700, "loss": 0.5912920236587524, "lr": 0.0003, "grad_norm": 0.16336441090518247, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78847753732, "tokens_per_second": 153392.51085559066, "data_wait_s": 1.1928001185879111e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 538800, "loss": 0.588059663772583, "lr": 0.0003, "grad_norm": 0.1442629493010716, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78862425442, "tokens_per_second": 153281.08856180214, "data_wait_s": 1.3524999303626828e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:baba0bbbb0af4d5ef69dcc7cb0cce95a25d8be7e03c024b84f298c2484bb3ef0
-size 291897
+oid sha256:f3f99996a999f3d02b911c49198236c9e0e907d63cfe63319e1c4684a90bd024
+size 294015