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": 77102.74731980654,
-      "required_pages_per_second_upper_estimate": 0.005334146962606025
+      "required_token_positions_per_second": 76923.80646283797,
+      "required_pages_per_second_upper_estimate": 0.005321767418920947
     },
     {
       "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": 9637.843414975818,
-      "required_pages_per_second_upper_estimate": 0.004035163776793712
+      "required_token_positions_per_second": 9615.475807854746,
+      "required_pages_per_second_upper_estimate": 0.0040257989267808705
     },
     {
       "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": 19275.686829951635,
-      "required_pages_per_second_upper_estimate": 0.004014617232633235
+      "required_token_positions_per_second": 19230.95161570949,
+      "required_pages_per_second_upper_estimate": 0.004005300067253532
     },
     {
       "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": 28913.530244927453,
-      "required_pages_per_second_upper_estimate": 0.0017333219897789924
+      "required_token_positions_per_second": 28846.427423564237,
+      "required_pages_per_second_upper_estimate": 0.001729299278098343
     },
     {
       "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": 28913.530244927453,
-      "required_pages_per_second_upper_estimate": 0.002247899799364061
+      "required_token_positions_per_second": 28846.427423564237,
+      "required_pages_per_second_upper_estimate": 0.0022426828501571888
     },
     {
       "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": 28913.530244927453,
-      "required_pages_per_second_upper_estimate": 0.3708717210519036
+      "required_token_positions_per_second": 28846.427423564237,
+      "required_pages_per_second_upper_estimate": 0.37001099810885235
     }
   ],
   "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": 0.5126764615936565,
-    "producer_p95_seconds": 0.2570298959981301,
-    "gpu_update_seconds": 0.9562305178853254,
-    "producer_batches_per_second": 1.9505479086976152,
-    "gpu_batches_per_second": 1.0457729399930358,
-    "producer_headroom": 1.8651734368940591,
+    "producer_batch_seconds": 0.08907482474214135,
+    "producer_p95_seconds": 0.2356768129975535,
+    "gpu_update_seconds": 0.9584549099974419,
+    "producer_batches_per_second": 11.226516615607771,
+    "gpu_batches_per_second": 1.0433458992897944,
+    "producer_headroom": 10.760109972397132,
     "sustainable": true,
-    "buffer_seconds": 60.242522626775504,
+    "buffer_seconds": 60.38265932983884,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 374138.9671109641,
-    "upload_bytes_per_second": 5712782.455975738,
+    "generation_bytes_per_second": 373852.41555562854,
+    "upload_bytes_per_second": 5671637.243297619,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.28151283261006,
+    "estimated_full_shards_per_hour": 80.22002553941385,
     "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": 1.0808812039995246,
-      "bytes": 6459806,
-      "pages": 30,
-      "shards": 3
-    },
     {
       "seconds": 1.1259296549997089,
       "bytes": 5894764,
@@ -562,6 +556,12 @@
       "bytes": 11740969,
       "pages": 33,
       "shards": 3
+    },
+    {
+      "seconds": 0.9664811299990106,
+      "bytes": 2719675,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -172,3 +172,4 @@
 {"step": 414300, "loss": 0.5566993951797485, "lr": 0.0003, "grad_norm": 0.14471358873226575, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60642563852, "tokens_per_second": 151106.99731864664, "data_wait_s": 1.5335001080529764e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 414400, "loss": 0.5966958999633789, "lr": 0.0003, "grad_norm": 0.15678041712656168, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60657135585, "tokens_per_second": 138509.9481808908, "data_wait_s": 1.4477998774964362e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 414500, "loss": 0.5783288478851318, "lr": 0.0003, "grad_norm": 0.1487296508465227, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60671859246, "tokens_per_second": 152432.4361111865, "data_wait_s": 1.1836000339826569e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 414600, "loss": 0.58902907371521, "lr": 0.0003, "grad_norm": 0.16139798390789045, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60686394448, "tokens_per_second": 149788.42157578593, "data_wait_s": 6.973001291044056e-06, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c55d372296ed44b85b0dfd0f2ad5b77cbb171697aea7211f6d0c977b8b2181b4
-size 371046
+oid sha256:656c119c44b30e92242ba2bac1db2a57f0dccb229c5befef63cb180b6465d903
+size 373164