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": 76193.79014935586,
-      "required_pages_per_second_upper_estimate": 0.005271263196483038
+      "required_token_positions_per_second": 75965.91376969202,
+      "required_pages_per_second_upper_estimate": 0.005255498179791845
     },
     {
       "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": 9524.223768669483,
-      "required_pages_per_second_upper_estimate": 0.003987593603533268
+      "required_token_positions_per_second": 9495.739221211503,
+      "required_pages_per_second_upper_estimate": 0.0039756677183376025
     },
     {
       "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": 19048.447537338965,
-      "required_pages_per_second_upper_estimate": 0.0039672892806851545
+      "required_token_positions_per_second": 18991.478442423006,
+      "required_pages_per_second_upper_estimate": 0.0039554241205902235
     },
     {
       "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": 28572.671306008448,
-      "required_pages_per_second_upper_estimate": 0.001712888016852262
+      "required_token_positions_per_second": 28487.217663634507,
+      "required_pages_per_second_upper_estimate": 0.0017077652014721011
     },
     {
       "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": 28572.671306008448,
-      "required_pages_per_second_upper_estimate": 0.0022213995161431315
+      "required_token_positions_per_second": 28487.217663634507,
+      "required_pages_per_second_upper_estimate": 0.002214755871319407
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28572.671306008448,
-      "required_pages_per_second_upper_estimate": 0.36649954856926475
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28487.217663634507,
+      "required_pages_per_second_upper_estimate": 0.3654034409978644
     }
   ],
   "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": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.24413314516391438,
-    "producer_p95_seconds": 0.22665722600140725,
-    "gpu_update_seconds": 0.9676379119017128,
-    "producer_batches_per_second": 4.096125494670485,
-    "gpu_batches_per_second": 1.0334444193434769,
-    "producer_headroom": 3.963566320550318,
+    "producer_batch_seconds": 0.09051832566393614,
+    "producer_p95_seconds": 0.21323752799798967,
+    "gpu_update_seconds": 0.9705405535372514,
+    "producer_batches_per_second": 11.047486712388617,
+    "gpu_batches_per_second": 1.0303536481349287,
+    "producer_headroom": 10.722033869037078,
     "sustainable": true,
-    "buffer_seconds": 60.961188449807906,
+    "buffer_seconds": 60.17351431930959,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 364298.241136121,
-    "upload_bytes_per_second": 5671637.243297619,
+    "generation_bytes_per_second": 404756.39109693415,
+    "upload_bytes_per_second": 5735219.0938478075,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.16992211878511,
+    "estimated_full_shards_per_hour": 86.85129928284663,
     "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": 1.1259296549997089,
-      "bytes": 5894764,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 0.93207276099929,
       "bytes": 6831941,
@@ -562,6 +556,12 @@
       "bytes": 2719675,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.2011403209980926,
+      "bytes": 11095830,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -174,3 +174,4 @@
 {"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}
 {"step": 414700, "loss": 0.5603664517402649, "lr": 0.0003, "grad_norm": 0.15124397056899685, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60701168716, "tokens_per_second": 153337.75768063735, "data_wait_s": 1.0086998372571543e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 414800, "loss": 0.5885172486305237, "lr": 0.0003, "grad_norm": 0.15518945816911042, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60715829472, "tokens_per_second": 152416.87492269906, "data_wait_s": 1.551599780214019e-05, "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:ef841ab58f4fd0c8e417dba76458b02ec5ed448fff9a792329ba502810d15d50
-size 375282
+oid sha256:be9127520e5c83b1568936371dcf60b5cd4653d88e6c8edae90d5968b1ae8190
+size 377400