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": 77188.50327254727,
-      "required_pages_per_second_upper_estimate": 0.005340079758397853
+      "required_token_positions_per_second": 77127.72751357965,
+      "required_pages_per_second_upper_estimate": 0.005335875150373284
     },
     {
       "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": 9648.562909068409,
-      "required_pages_per_second_upper_estimate": 0.004039651805121789
+      "required_token_positions_per_second": 9640.965939197456,
+      "required_pages_per_second_upper_estimate": 0.004036471112479592
     },
     {
       "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": 19297.125818136818,
-      "required_pages_per_second_upper_estimate": 0.004019082408488071
+      "required_token_positions_per_second": 19281.931878394913,
+      "required_pages_per_second_upper_estimate": 0.004015917911530968
     },
     {
       "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": 28945.688727205226,
-      "required_pages_per_second_upper_estimate": 0.0017352498416883856
+      "required_token_positions_per_second": 28922.89781759237,
+      "required_pages_per_second_upper_estimate": 0.0017338835614567989
     },
     {
       "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": 28945.688727205226,
-      "required_pages_per_second_upper_estimate": 0.002250399979910943
+      "required_token_positions_per_second": 28922.89781759237,
+      "required_pages_per_second_upper_estimate": 0.0022486280869351604
     },
     {
       "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": 28945.688727205226,
-      "required_pages_per_second_upper_estimate": 0.3712842155334748
+      "required_token_positions_per_second": 28922.89781759237,
+      "required_pages_per_second_upper_estimate": 0.3709918782159332
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.09123250236723379,
-    "producer_p95_seconds": 0.32287491299939575,
-    "gpu_update_seconds": 0.9551681516568798,
-    "producer_batches_per_second": 10.961005935962913,
-    "gpu_batches_per_second": 1.0469360795430132,
-    "producer_headroom": 10.469603780153783,
+    "producer_batch_seconds": 0.13269625375806982,
+    "producer_p95_seconds": 0.3207643409987213,
+    "gpu_update_seconds": 0.9559208131345363,
+    "producer_batches_per_second": 7.536007774742366,
+    "gpu_batches_per_second": 1.0461117555552795,
+    "producer_headroom": 7.203826679819909,
     "sustainable": true,
-    "buffer_seconds": 60.17559355438343,
+    "buffer_seconds": 60.22301122747579,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 445811.8229312471,
-    "upload_bytes_per_second": 4781441.359220344,
+    "generation_bytes_per_second": 475885.12390647153,
+    "upload_bytes_per_second": 4850469.2760212105,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 95.66083923295079,
+    "estimated_full_shards_per_hour": 102.11386955161676,
     "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": 1.4462293939977826,
-      "bytes": 810671,
-      "pages": 26,
-      "shards": 1
-    },
     {
       "seconds": 1.2543598109987215,
       "bytes": 12226155,
@@ -562,6 +556,12 @@
       "bytes": 6310823,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.2331648629988194,
+      "bytes": 6006485,
+      "pages": 30,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -344,3 +344,4 @@
 {"step": 959400, "loss": 0.5365502238273621, "lr": 3e-05, "grad_norm": 0.347566418884559, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140430128981, "tokens_per_second": 152366.0358528447, "data_wait_s": 1.1727002856787294e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 959500, "loss": 0.4938940405845642, "lr": 3e-05, "grad_norm": 0.3375864812509123, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140444877119, "tokens_per_second": 153630.99446575344, "data_wait_s": 1.200000406242907e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 959600, "loss": 0.5301459431648254, "lr": 3e-05, "grad_norm": 0.344852297676844, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140459388377, "tokens_per_second": 150807.61525164032, "data_wait_s": 1.1290001566521823e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 959700, "loss": 0.4841211438179016, "lr": 3e-05, "grad_norm": 0.34537294225829357, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140473890672, "tokens_per_second": 150508.60422501882, "data_wait_s": 1.7505000869277865e-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 sha256:9368b38cc223b63ff72128091aa73a8e812d6a9489a0dbfb2feae89c9ea72a38
-size 737691
+oid sha256:af20071fdf1018e57e6f1b7fbcf744555992b3a1f4dda5b1ed18d43250870193
+size 739809