Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75807.85978511196,
-      "required_pages_per_second_upper_estimate": 0.006570833152909446
+      "required_token_positions_per_second": 75650.37661255106,
+      "required_pages_per_second_upper_estimate": 0.00655718291064932
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9475.982473138994,
-      "required_pages_per_second_upper_estimate": 0.0037801110870986894
+      "required_token_positions_per_second": 9456.297076568882,
+      "required_pages_per_second_upper_estimate": 0.003772258288083965
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18951.96494627799,
-      "required_pages_per_second_upper_estimate": 0.004384203547130273
+      "required_token_positions_per_second": 18912.594153137765,
+      "required_pages_per_second_upper_estimate": 0.004375095807039576
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28427.947419416985,
-      "required_pages_per_second_upper_estimate": 0.0017311128479292975
+      "required_token_positions_per_second": 28368.89122970665,
+      "required_pages_per_second_upper_estimate": 0.0017275166358198192
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28427.947419416985,
-      "required_pages_per_second_upper_estimate": 0.0021884620859417905
+      "required_token_positions_per_second": 28368.89122970665,
+      "required_pages_per_second_upper_estimate": 0.002183915776979893
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28427.947419416985,
-      "required_pages_per_second_upper_estimate": 0.3529799647294658
+      "required_token_positions_per_second": 28368.89122970665,
+      "required_pages_per_second_upper_estimate": 0.352246684501616
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.12273629364065641,
-    "producer_p95_seconds": 0.26120527800230775,
-    "gpu_update_seconds": 0.9725640614178053,
-    "producer_batches_per_second": 8.147549272815501,
-    "gpu_batches_per_second": 1.0282099037694221,
-    "producer_headroom": 7.92401361137113,
+    "producer_batch_seconds": 0.323214989546841,
+    "producer_p95_seconds": 0.2839081140009512,
+    "gpu_update_seconds": 0.9745886709540568,
+    "producer_batches_per_second": 3.0939159146116206,
+    "gpu_batches_per_second": 1.0260739015374223,
+    "producer_headroom": 3.0152953992649443,
     "sustainable": true,
-    "buffer_seconds": 60.298971807903925,
+    "buffer_seconds": 60.42449759915152,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 303202.6017693626,
-    "upload_bytes_per_second": 5459608.194913292,
+    "generation_bytes_per_second": 292004.2333301237,
+    "upload_bytes_per_second": 5499703.013765946,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.06022014437349,
+    "estimated_full_shards_per_hour": 62.657310961988294,
     "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": 0.9390472680024686,
-      "bytes": 6039556,
-      "pages": 27,
-      "shards": 2
-    },
     {
       "seconds": 1.1019230559977586,
       "bytes": 1147220,
@@ -562,6 +556,12 @@
       "bytes": 2731200,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.413960779998888,
+      "bytes": 11724293,
+      "pages": 29,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -234,3 +234,4 @@
 {"step": 338400, "loss": 0.5691492799669504, "lr": 0.0003, "grad_norm": 0.14570167071391565, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49533027987, "tokens_per_second": 149911.24673057065, "data_wait_s": 8.025999704841524e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 338500, "loss": 0.5700047565624118, "lr": 0.0003, "grad_norm": 0.1598025247832868, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49547452233, "tokens_per_second": 147611.6312319545, "data_wait_s": 1.3952001609141007e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 338600, "loss": 0.6005110405385494, "lr": 0.0003, "grad_norm": 0.19402572089345949, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49561984563, "tokens_per_second": 147040.83739049642, "data_wait_s": 1.6848000086611137e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 338700, "loss": 0.6105085415765643, "lr": 0.0003, "grad_norm": 0.16156311645911683, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49576486868, "tokens_per_second": 135554.30426191163, "data_wait_s": 1.5159999747993425e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0176025f6f981d38c0e0e654463515d215550769ef3d31a67b3d5af7334298d4
-size 503145
+oid sha256:bb51eeab64d21446893e31a36d5b4caf548ac02307e76869f4777b2b58313b4e
+size 505263