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": 77156.91683157929,
-      "required_pages_per_second_upper_estimate": 0.0053378945351210474
+      "required_token_positions_per_second": 77385.33280119981,
+      "required_pages_per_second_upper_estimate": 0.005353696881897464
     },
     {
       "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": 9644.614603947412,
-      "required_pages_per_second_upper_estimate": 0.004037998732217614
+      "required_token_positions_per_second": 9673.166600149976,
+      "required_pages_per_second_upper_estimate": 0.004049952856794148
     },
     {
       "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": 19289.229207894823,
-      "required_pages_per_second_upper_estimate": 0.004017437752822279
+      "required_token_positions_per_second": 19346.333200299952,
+      "required_pages_per_second_upper_estimate": 0.0040293310085067185
     },
     {
       "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": 28933.843811842235,
-      "required_pages_per_second_upper_estimate": 0.0017345397570985128
+      "required_token_positions_per_second": 29019.49980044993,
+      "required_pages_per_second_upper_estimate": 0.0017396747028264224
     },
     {
       "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": 28933.843811842235,
-      "required_pages_per_second_upper_estimate": 0.002249479089841747
+      "required_token_positions_per_second": 29019.49980044993,
+      "required_pages_per_second_upper_estimate": 0.0022561384661951185
     },
     {
       "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": 28933.843811842235,
-      "required_pages_per_second_upper_estimate": 0.3711322816772775
+      "required_token_positions_per_second": 29019.49980044993,
+      "required_pages_per_second_upper_estimate": 0.372230984728902
     }
   ],
   "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": 0.1362607551951669,
-    "producer_p95_seconds": 0.3206976119981846,
-    "gpu_update_seconds": 0.9555591776811916,
-    "producer_batches_per_second": 7.338870231327395,
-    "gpu_batches_per_second": 1.0465076610185993,
-    "producer_headroom": 7.012724803356182,
+    "producer_batch_seconds": 0.09012441237530311,
+    "producer_p95_seconds": 0.31713193299947307,
+    "gpu_update_seconds": 0.9527386822694764,
+    "producer_batches_per_second": 11.095772761721008,
+    "gpu_batches_per_second": 1.049605750884329,
+    "producer_headroom": 10.571371919763623,
     "sustainable": true,
-    "buffer_seconds": 60.200228193915066,
+    "buffer_seconds": 60.022536982977016,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 157,
+    "upload_pages_per_commit": 166,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 687626.2334901626,
-    "upload_bytes_per_second": 5435542.839569905,
+    "generation_bytes_per_second": 676267.3357613685,
+    "upload_bytes_per_second": 5430522.109190463,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.77429129375773,
+    "estimated_full_shards_per_hour": 72.55561377826115,
     "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": 1.2462927969972952,
-      "bytes": 6148998,
-      "pages": 33,
-      "shards": 2
-    },
     {
       "seconds": 1.1124823669997568,
       "bytes": 8182028,
@@ -562,6 +556,12 @@
       "bytes": 2748829,
       "pages": 38,
       "shards": 2
+    },
+    {
+      "seconds": 1.300822404999053,
+      "bytes": 6046016,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -352,3 +352,4 @@
 {"step": 680200, "loss": 0.5280165672302246, "lr": 0.0001, "grad_norm": 0.2234420342974073, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99563650571, "tokens_per_second": 153044.18241708132, "data_wait_s": 1.1339005141053349e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 680300, "loss": 0.5367920994758606, "lr": 0.0001, "grad_norm": 0.1894354740101249, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99578384978, "tokens_per_second": 153525.13770561607, "data_wait_s": 1.1192001693416387e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 680400, "loss": 0.6108964681625366, "lr": 0.0001, "grad_norm": 0.24687034693302684, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99592982197, "tokens_per_second": 152342.38607546993, "data_wait_s": 9.714000043459237e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 680500, "loss": 0.592078447341919, "lr": 0.0001, "grad_norm": 0.2525708030391718, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99607513653, "tokens_per_second": 150977.59807706944, "data_wait_s": 1.757699646987021e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4162c93fdfbadb9039adcd04f5f28f850d00a5c477a16fa2c8b7d12a0910dac6
-size 755418
+oid sha256:c0081919fae953df2a0b4f7154a6e734eb8ec88a49f2bcc61ea745b796f0303c
+size 757536