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": 76629.06178479652,
-      "required_pages_per_second_upper_estimate": 0.005301376298192164
+      "required_token_positions_per_second": 77141.68126538243,
+      "required_pages_per_second_upper_estimate": 0.005336840503300165
     },
     {
       "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": 9578.632723099565,
-      "required_pages_per_second_upper_estimate": 0.004010373496564974
+      "required_token_positions_per_second": 9642.710158172804,
+      "required_pages_per_second_upper_estimate": 0.004037201380541136
     },
     {
       "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": 19157.26544619913,
-      "required_pages_per_second_upper_estimate": 0.00398995318137949
+      "required_token_positions_per_second": 19285.420316345608,
+      "required_pages_per_second_upper_estimate": 0.004016644461159803
     },
     {
       "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": 28735.898169298696,
-      "required_pages_per_second_upper_estimate": 0.0017226732180735233
+      "required_token_positions_per_second": 28928.130474518413,
+      "required_pages_per_second_upper_estimate": 0.0017341972512497107
     },
     {
       "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": 28735.898169298696,
-      "required_pages_per_second_upper_estimate": 0.002234089686804853
+      "required_token_positions_per_second": 28928.130474518413,
+      "required_pages_per_second_upper_estimate": 0.002249034902995135
     },
     {
       "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": 28735.898169298696,
-      "required_pages_per_second_upper_estimate": 0.3685932475122009
+      "required_token_positions_per_second": 28928.130474518413,
+      "required_pages_per_second_upper_estimate": 0.37105899712059126
     }
   ],
   "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": 0.43347808859377324,
-    "producer_p95_seconds": 0.21979521100001875,
-    "gpu_update_seconds": 0.9621414941377749,
-    "producer_batches_per_second": 2.3069216791189033,
-    "gpu_batches_per_second": 1.0393481687391022,
-    "producer_headroom": 2.219585071206286,
+    "producer_batch_seconds": 0.08165683560164894,
+    "producer_p95_seconds": 0.22152489099971717,
+    "gpu_update_seconds": 0.9557479016611174,
+    "producer_batches_per_second": 12.246372182218222,
+    "gpu_batches_per_second": 1.0463010154267367,
+    "producer_headroom": 11.704444516116144,
     "sustainable": true,
-    "buffer_seconds": 60.61491413067982,
+    "buffer_seconds": 60.2121178046504,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 335831.1600795291,
-    "upload_bytes_per_second": 5300682.556384319,
+    "generation_bytes_per_second": 350240.29239175934,
+    "upload_bytes_per_second": 5323656.211749268,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.06154920377163,
+    "estimated_full_shards_per_hour": 75.15341357054315,
     "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": 1.1031485080002312,
-      "bytes": 6510962,
-      "pages": 32,
-      "shards": 3
-    },
     {
       "seconds": 1.2347332800000004,
       "bytes": 6132473,
@@ -562,6 +556,12 @@
       "bytes": 5610482,
       "pages": 19,
       "shards": 2
+    },
+    {
+      "seconds": 1.425927022999531,
+      "bytes": 10079783,
+      "pages": 46,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -168,3 +168,4 @@
 {"step": 573900, "loss": 0.594629168510437, "lr": 0.0003, "grad_norm": 0.15814314169985533, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84001990587, "tokens_per_second": 153462.2708640967, "data_wait_s": 1.4851000742055476e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 574000, "loss": 0.5856558084487915, "lr": 0.0003, "grad_norm": 0.1522044536434727, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84016768933, "tokens_per_second": 154722.74971176902, "data_wait_s": 8.637001883471385e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 574100, "loss": 0.5379246473312378, "lr": 0.0003, "grad_norm": 0.18402141610751724, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84031449939, "tokens_per_second": 131570.80657106868, "data_wait_s": 1.5387999155791476e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 574200, "loss": 0.5574430227279663, "lr": 0.0003, "grad_norm": 0.17452285139223, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84046065438, "tokens_per_second": 151739.14838743364, "data_wait_s": 1.6810001397971064e-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:e60e73ec24691c337178f874de84cc9a6ed7ff7b349e0901394a68131962232f
-size 362574
+oid sha256:942446c3e63cb22cbfce8dd2f5cac312e4cff6590c14d48c756b7a233c69d81b
+size 364692