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": 77068.27212078709,
-      "required_pages_per_second_upper_estimate": 0.005331761888344376
+      "required_token_positions_per_second": 76785.10543538093,
+      "required_pages_per_second_upper_estimate": 0.005312171760010746
     },
     {
       "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": 9633.534015098387,
-      "required_pages_per_second_upper_estimate": 0.004033359521055535
+      "required_token_positions_per_second": 9598.138179422616,
+      "required_pages_per_second_upper_estimate": 0.00401854002380719
     },
     {
       "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": 19267.068030196773,
-      "required_pages_per_second_upper_estimate": 0.0040128221639373325
+      "required_token_positions_per_second": 19196.276358845233,
+      "required_pages_per_second_upper_estimate": 0.003998078125696723
     },
     {
       "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": 28900.602045295156,
-      "required_pages_per_second_upper_estimate": 0.001732546963951254
+      "required_token_positions_per_second": 28794.41453826785,
+      "required_pages_per_second_upper_estimate": 0.001726181185043902
     },
     {
       "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": 28900.602045295156,
-      "required_pages_per_second_upper_estimate": 0.00224689468870779
+      "required_token_positions_per_second": 28794.41453826785,
+      "required_pages_per_second_upper_estimate": 0.002238639077105899
     },
     {
       "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": 28900.602045295156,
-      "required_pages_per_second_upper_estimate": 0.3707058919882397
+      "required_token_positions_per_second": 28794.41453826785,
+      "required_pages_per_second_upper_estimate": 0.3693438326633554
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.05376410415625088,
-    "producer_p95_seconds": 0.16859828399992693,
-    "gpu_update_seconds": 0.9566582715705373,
-    "producer_batches_per_second": 18.5997705289345,
-    "gpu_batches_per_second": 1.0453053401799464,
-    "producer_headroom": 17.793624325819096,
+    "producer_batch_seconds": 0.36178068457811463,
+    "producer_p95_seconds": 0.18755171299926587,
+    "gpu_update_seconds": 0.9601862181728243,
+    "producer_batches_per_second": 2.764105555182239,
+    "gpu_batches_per_second": 1.041464646204711,
+    "producer_headroom": 2.654056059660929,
     "sustainable": true,
-    "buffer_seconds": 60.26947110894385,
+    "buffer_seconds": 60.491731744887936,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 299712.93419618456,
-    "upload_bytes_per_second": 5407182.046672796,
+    "generation_bytes_per_second": 288658.2123737393,
+    "upload_bytes_per_second": 5419577.50955878,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.31141871847298,
+    "estimated_full_shards_per_hour": 61.93933275612959,
     "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": [
@@ -538,6 +538,12 @@
       "bytes": 1220170,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 0.9136476189996756,
+      "bytes": 5828433,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -71,3 +71,4 @@
 {"step": 564200, "loss": 0.6458748579025269, "lr": 0.0003, "grad_norm": 0.159881705543696, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82581368181, "tokens_per_second": 151678.36277086398, "data_wait_s": 1.4647000170953106e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 564300, "loss": 0.5641994476318359, "lr": 0.0003, "grad_norm": 0.15571673443048747, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82596009954, "tokens_per_second": 152336.7328731782, "data_wait_s": 1.3189999663154595e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 564400, "loss": 0.5654087662696838, "lr": 0.0003, "grad_norm": 0.165348709344066, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82610859212, "tokens_per_second": 154824.6561223058, "data_wait_s": 7.89299974712776e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 564500, "loss": 0.6321799159049988, "lr": 0.0003, "grad_norm": 0.16649485057799607, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 82625466448, "tokens_per_second": 152091.64661745608, "data_wait_s": 1.189700014947448e-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 sha256:52bec16e59f923606467fd280b48111aab68fb8941bd450b8a1ca2f45e4b90f9
-size 155562
+oid sha256:d0e98856543863bca34cfe804ea995caaadc38eaec75677eb38b043249802ef9
+size 157680