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": 77525.63046015309,
-      "required_pages_per_second_upper_estimate": 0.005363402999479271
+      "required_token_positions_per_second": 77215.51033368144,
+      "required_pages_per_second_upper_estimate": 0.00534194816955213
     },
     {
       "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": 9690.703807519136,
-      "required_pages_per_second_upper_estimate": 0.00405729531930108
+      "required_token_positions_per_second": 9651.93879171018,
+      "required_pages_per_second_upper_estimate": 0.004041065216687453
     },
     {
       "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": 19381.40761503827,
-      "required_pages_per_second_upper_estimate": 0.0040366360841221915
+      "required_token_positions_per_second": 19303.87758342036,
+      "required_pages_per_second_upper_estimate": 0.004020488623140608
     },
     {
       "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": 29072.11142255741,
-      "required_pages_per_second_upper_estimate": 0.0017428286892384702
+      "required_token_positions_per_second": 28955.816375130536,
+      "required_pages_per_second_upper_estimate": 0.001735856978717485
     },
     {
       "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": 29072.11142255741,
-      "required_pages_per_second_upper_estimate": 0.002260228788399906
+      "required_token_positions_per_second": 28955.816375130536,
+      "required_pages_per_second_upper_estimate": 0.0022511873599903205
     },
     {
       "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": 29072.11142255741,
-      "required_pages_per_second_upper_estimate": 0.37290583012733813
+      "required_token_positions_per_second": 28955.816375130536,
+      "required_pages_per_second_upper_estimate": 0.37141412212683955
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.24235251214076925,
-    "producer_p95_seconds": 0.34582703599880915,
-    "gpu_update_seconds": 0.9510145169073471,
-    "producer_batches_per_second": 4.126220896852742,
-    "gpu_batches_per_second": 1.0515086596700451,
-    "producer_headroom": 3.924095972873411,
+    "producer_batch_seconds": 0.10153320464866056,
+    "producer_p95_seconds": 0.35486642199975904,
+    "gpu_update_seconds": 0.9548340700124832,
+    "producer_batches_per_second": 9.848994754576498,
+    "gpu_batches_per_second": 1.0473023862532747,
+    "producer_headroom": 9.404155747043877,
     "sustainable": true,
-    "buffer_seconds": 60.86492908207021,
+    "buffer_seconds": 60.154546410786445,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 422770.81334018405,
-    "upload_bytes_per_second": 4925400.543857028,
+    "generation_bytes_per_second": 423278.3939063002,
+    "upload_bytes_per_second": 5041480.47465339,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 90.71677494196072,
+    "estimated_full_shards_per_hour": 90.82568991557841,
     "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": 1.3606159429982654,
-      "bytes": 2419629,
-      "pages": 29,
-      "shards": 2
-    },
     {
       "seconds": 1.757866293999541,
       "bytes": 6331179,
@@ -562,6 +556,12 @@
       "bytes": 5081198,
       "pages": 30,
       "shards": 2
+    },
+    {
+      "seconds": 1.6100998589972733,
+      "bytes": 13518430,
+      "pages": 43,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -410,3 +410,4 @@
 {"step": 966000, "loss": 0.5483071208000183, "lr": 3e-05, "grad_norm": 0.3376321497210277, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141395850135, "tokens_per_second": 151453.7854968563, "data_wait_s": 1.7395999748259783e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 966100, "loss": 0.5369018912315369, "lr": 3e-05, "grad_norm": 0.37190364193231235, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141410572187, "tokens_per_second": 153852.86583037066, "data_wait_s": 7.890994311310351e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 966200, "loss": 0.546553909778595, "lr": 3e-05, "grad_norm": 0.3533076130064056, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141425317778, "tokens_per_second": 153513.28357493415, "data_wait_s": 1.846899976953864e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 966300, "loss": 0.5306552052497864, "lr": 3e-05, "grad_norm": 0.355788559411236, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141440034338, "tokens_per_second": 153460.48339376223, "data_wait_s": 1.8106999050360173e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:eba9323d731ac50076d64de85ed1786709f8de8a718309703652686ebb8e5941
-size 879045
+oid sha256:61d9d7d330670bc231bd13b91b5fda3c493ae09df80496f702b3d2f3c54281e5
+size 881163