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": 77564.49085135169,
-      "required_pages_per_second_upper_estimate": 0.00536609145151092
+      "required_token_positions_per_second": 77215.40583494017,
+      "required_pages_per_second_upper_estimate": 0.005341940940086739
     },
     {
       "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": 9695.56135641896,
-      "required_pages_per_second_upper_estimate": 0.004059329073588282
+      "required_token_positions_per_second": 9651.925729367522,
+      "required_pages_per_second_upper_estimate": 0.004041059747757355
     },
     {
       "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": 19391.12271283792,
-      "required_pages_per_second_upper_estimate": 0.0040386594827895
+      "required_token_positions_per_second": 19303.851458735044,
+      "required_pages_per_second_upper_estimate": 0.004020483182057611
     },
     {
       "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": 29086.68406925688,
-      "required_pages_per_second_upper_estimate": 0.0017437022971569605
+      "required_token_positions_per_second": 28955.777188102566,
+      "required_pages_per_second_upper_estimate": 0.0017358546295149933
     },
     {
       "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": 29086.68406925688,
-      "required_pages_per_second_upper_estimate": 0.002261361747582491
+      "required_token_positions_per_second": 28955.777188102566,
+      "required_pages_per_second_upper_estimate": 0.002251184313371261
     },
     {
       "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": 29086.68406925688,
-      "required_pages_per_second_upper_estimate": 0.37309275239231
+      "required_token_positions_per_second": 28955.777188102566,
+      "required_pages_per_second_upper_estimate": 0.37141361947772045
     }
   ],
   "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": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.06650559086719454,
-    "producer_p95_seconds": 0.16056235100040794,
-    "gpu_update_seconds": 0.9505380515073049,
-    "producer_batches_per_second": 15.036329832734014,
-    "gpu_batches_per_second": 1.0520357374586544,
-    "producer_headroom": 14.292603661028147,
+    "producer_batch_seconds": 0.06807644087503917,
+    "producer_p95_seconds": 0.1615099470000132,
+    "gpu_update_seconds": 0.9548353622281667,
+    "producer_batches_per_second": 14.689369584341165,
+    "gpu_batches_per_second": 1.0473009688983856,
+    "producer_headroom": 14.02592952796781,
     "sustainable": true,
-    "buffer_seconds": 60.83443529646751,
+    "buffer_seconds": 60.1546278203745,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 397019.83099778974,
-    "upload_bytes_per_second": 5683527.506870925,
+    "generation_bytes_per_second": 349733.4618013243,
+    "upload_bytes_per_second": 5730174.961226118,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.19121358347196,
+    "estimated_full_shards_per_hour": 75.04465952424809,
     "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": [
@@ -460,6 +460,12 @@
       "bytes": 5739919,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.3039567450005052,
+      "bytes": 9869427,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -54,3 +54,4 @@
 {"step": 562500, "loss": 0.6309000253677368, "lr": 0.0003, "grad_norm": 0.15265571888679594, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82332737079, "tokens_per_second": 153379.97200531844, "data_wait_s": 1.0342000678065233e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 562600, "loss": 0.607856810092926, "lr": 0.0003, "grad_norm": 0.16134249536027215, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82347508268, "tokens_per_second": 153913.72477537685, "data_wait_s": 1.625200002308702e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 562700, "loss": 0.5789791345596313, "lr": 0.0003, "grad_norm": 0.1317799438170744, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82362214380, "tokens_per_second": 142841.35138604647, "data_wait_s": 1.5511999663431197e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 562800, "loss": 0.5476948618888855, "lr": 0.0003, "grad_norm": 0.15059537639775994, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82376740089, "tokens_per_second": 151489.53378136404, "data_wait_s": 1.3345000297704246e-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:f19bfe2125472aaaa88382dca8f3c788a750fc067ad42b5655cffa7809a229cd
-size 119556
+oid sha256:0e7901205a66446d7e86a850000af5845bda3528d6de429a2ad62edafe776da2
+size 121674